A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

SegFormer B3 ADE20K — 150-Class Scene Segmentation

Label image regions such as walls, roads, furniture, and vegetation. Returns pixel masks for predicted categories from 150 indoor and outdoor scene classes, rather than separate identities for each object.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Label image regions such as walls, roads, furniture, and vegetation. Returns pixel masks for predicted categories from 150 indoor and outdoor scene classes, rather than separate identities for each object.

Created by
NVIDIA
Model released
Unknown
Source repository created
Mar 2, 2022
Model weights · Q4F16
47.9 MB

Best for

  • Non-commercial research and evaluation of semantic scene segmentation.
  • Educational visualization of ADE20K scene classes.
  • Prototype browser tools that need coarse per-pixel scene labels.

Keep in mind

  • The NVIDIA SegFormer license restricts use to non-commercial research or evaluation and requires preserving its license and notices.
  • The fixed ADE20K taxonomy is incomplete and merges visually or functionally distinct objects.
  • Predictions can fail on small objects, fine boundaries, unusual viewpoints, low light, occlusion, or domains unlike ADE20K.
  • Masks are resized to the source image and can lose boundary detail after the model's 512×512 preprocessing.
  • Output must not be used for surveillance, safety-critical navigation, medical analysis, or other high-impact automated decisions.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
nvidia/segformer-b3-finetuned-ade-512-512
Release verification
Unknown
Release evidence
Exact original checkpoint public release date has not been verified. Repository creation, training, paper, artifact conversion and update dates are not release evidence.
Source updated
Aug 6, 2022
Browser model files
onnx-community/segformer-b3-finetuned-ade-512-512-ONNX
Files updated
Aug 2, 2026
Artifact revision
748163478112c6185cd52530d7b133d96c8236be

Input

IMAGE

Sample result

SegFormer B3 ADE20K — 150-Class Scene Segmentation

Settings

Adds a self-contained SVG presentation image alongside the unchanged native model fields. When enabled, draws model predictions or presents generated media and text. Adds CPU postprocessing time and image bytes. It does not change model predictions.

Output

Your result will appear here

Add input or try the sample, then run the model.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

SegFormer B3 ADE20K — 150-Class Scene Segmentation

16 configurations tested
Passed16
Failed0
TestedSep 15, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Finite original-size binary masks with unique known labels; checked-in expected portrait scene labels and area ranges must match the pinned FP32 reference signature.

PrecisionDeviceResult
Q4F16GPU · WebGPUAcceleratedPassedModerate
Q4GPU · WebGPUAcceleratedPassedModerate
BNB4GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
UINT8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedVery slow
INT8GPU · WebGPUAcceleratedPassedVery slow
FP32CPU · WASMPassedSlow
Q4CPU · WASMPassedSlow
BNB4CPU · WASMPassedSlow
UINT8CPU · WASMPassedSlow
FP16CPU · WASMPassedVery slow
Q4F16CPU · WASMPassedVery slow
INT8CPU · WASMPassedVery slow
Q8CPU · WASMPassedVery slow
Test setup & sample details
Quality gate
Finite original-size binary masks with unique known labels; checked-in expected portrait scene labels and area ranges must match the pinned FP32 reference signature.
Tested
2026-09-15T14:19:59.145Z
Environment
Isolated headless Chrome on macOS; Transformers.js 4.2.0
Runtime
Transformers.js 4.2.0
Artifact revision
748163478112c6185cd52530d7b133d96c8236be
Fixture
{
  "file": "modnet-portrait-pexels-5965592.jpg",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "width": 1024,
  "height": 683,
  "source": "https://www.pexels.com/photo/woman-wearing-pink-turtleneck-sweater-5965592/",
  "license": "Pexels License",
  "licenseUrl": "https://www.pexels.com/license/",
  "preprocessing": "RGB decode through the pinned artifact's SegFormer image processor."
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 30140.9,
      "inferenceMs": 4752.3,
      "totalMs": 36212,
      "megapixelsPerSecond": 0.1471691601957464,
      "grade": "very-slow"
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 28649.6,
      "inferenceMs": 973.1,
      "totalMs": 30868,
      "megapixelsPerSecond": 0.7187257219240406,
      "grade": "slow"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 15895.6,
      "inferenceMs": 5107.5,
      "totalMs": 22349.899999976158,
      "megapixelsPerSecond": 0.13693431228585412,
      "grade": "very-slow"
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 14317.5,
      "inferenceMs": 1124.9,
      "totalMs": 16820.69999998808,
      "megapixelsPerSecond": 0.6217370432894428,
      "grade": "slow"
    }
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 10898.8,
      "inferenceMs": 9320.8,
      "totalMs": 21521.59999999404,
      "megapixelsPerSecond": 0.07503561926005338,
      "grade": "very-slow"
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 12340.8,
      "inferenceMs": 9439.8,
      "totalMs": 23167.29999998212,
      "megapixelsPerSecond": 0.07408970529027276,
      "grade": "very-slow"
    }
  },
  {
    "precision": "uint8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 11711.2,
      "inferenceMs": 4868.9,
      "totalMs": 17868,
      "megapixelsPerSecond": 0.1436447657580036,
      "grade": "very-slow"
    }
  },
  {
    "precision": "uint8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 10605.8,
      "inferenceMs": 4926.4,
      "totalMs": 16861.29999998212,
      "megapixelsPerSecond": 0.14196817148407637,
      "grade": "very-slow"
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 11630.7,
      "inferenceMs": 9322.1,
      "totalMs": 22232.79999998212,
      "megapixelsPerSecond": 0.07502515527621965,
      "grade": "very-slow"
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 11040.4,
      "inferenceMs": 9365.6,
      "totalMs": 21809.100000023842,
      "megapixelsPerSecond": 0.07467668916037895,
      "grade": "very-slow"
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 15844.1,
      "inferenceMs": 4798.4,
      "totalMs": 21937.90000000596,
      "megapixelsPerSecond": 0.14575525175040246,
      "grade": "very-slow"
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 19026.3,
      "inferenceMs": 961.1,
      "totalMs": 21425.59999999404,
      "megapixelsPerSecond": 0.7276995109815185,
      "grade": "slow"
    }
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 11557.5,
      "inferenceMs": 5203.8,
      "totalMs": 18118.59999999404,
      "megapixelsPerSecond": 0.13440024597378797,
      "grade": "very-slow"
    }
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 10842.6,
      "inferenceMs": 947.2,
      "totalMs": 13098.800000011921,
      "megapixelsPerSecond": 0.7383783783876712,
      "grade": "slow"
    }
  },
  {
    "precision": "bnb4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 18473.8,
      "inferenceMs": 4866.2,
      "totalMs": 24743,
      "megapixelsPerSecond": 0.1437244667300385,
      "grade": "very-slow"
    }
  },
  {
    "precision": "bnb4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "grade": "scene-segmentation-pass"
    },
    "runtime": {
      "loadMs": 15142.8,
      "inferenceMs": 970.2,
      "totalMs": 17421.70000001788,
      "megapixelsPerSecond": 0.7208740465750462,
      "grade": "slow"
    }
  }
]

SegFormer B0 · ADE20K (512×512)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
d3e5499fa8701ff0453ca940a8dfeae39b2f1504
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4699.9000000059605,
      "inferenceMs": 1181.4000000059605,
      "totalMs": 5881.300000011921,
      "megapixelsPerSecond": 0.5920027086477665,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4699.9000000059605,
      "inferenceMs": 486.09999999403954,
      "totalMs": 5186,
      "megapixelsPerSecond": 0.5392799835490935,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4699.9000000059605,
      "inferenceMs": 360.90000000596046,
      "totalMs": 5060.800000011921,
      "megapixelsPerSecond": 0.7263618730830439,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4543.4000000059605,
      "inferenceMs": 1268,
      "totalMs": 5811.4000000059605,
      "megapixelsPerSecond": 0.551570977917981,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4543.4000000059605,
      "inferenceMs": 852.5,
      "totalMs": 5395.9000000059605,
      "megapixelsPerSecond": 0.30750029325513195,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4543.4000000059605,
      "inferenceMs": 785.5999999940395,
      "totalMs": 5329,
      "megapixelsPerSecond": 0.33368635438135047,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4951.5999999940395,
      "inferenceMs": 1035.800000011921,
      "totalMs": 5987.4000000059605,
      "megapixelsPerSecond": 0.6752191542691164,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4951.5999999940395,
      "inferenceMs": 468.60000002384186,
      "totalMs": 5420.200000017881,
      "megapixelsPerSecond": 0.559419547560099,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4951.5999999940395,
      "inferenceMs": 349.59999999403954,
      "totalMs": 5301.199999988079,
      "megapixelsPerSecond": 0.7498398169464228,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5347.199999988079,
      "inferenceMs": 1411.9000000059605,
      "totalMs": 6759.0999999940395,
      "megapixelsPerSecond": 0.49535519512504245,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5347.199999988079,
      "inferenceMs": 900.9000000059605,
      "totalMs": 6248.0999999940395,
      "megapixelsPerSecond": 0.2909801309782058,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5347.199999988079,
      "inferenceMs": 764.8000000119209,
      "totalMs": 6112,
      "megapixelsPerSecond": 0.342761506270808,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4560.5999999940395,
      "inferenceMs": 1316.4000000059605,
      "totalMs": 5877,
      "megapixelsPerSecond": 0.5312914007876278,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4560.5999999940395,
      "inferenceMs": 873.0999999940395,
      "totalMs": 5433.699999988079,
      "megapixelsPerSecond": 0.3002451036556976,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4560.5999999940395,
      "inferenceMs": 732.6000000238419,
      "totalMs": 5293.200000017881,
      "megapixelsPerSecond": 0.35782691781527265,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "d3e5499fa8701ff0453ca940a8dfeae39b2f1504",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B0 · Cityscapes (1024×1024)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
64e537ca2bf6bf2afc13727ee6bb61cfca9e1048
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4728.399999976158,
      "inferenceMs": 631.5,
      "totalMs": 5359.899999976158,
      "megapixelsPerSecond": 1.1075091053048298,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4728.399999976158,
      "inferenceMs": 323.5,
      "totalMs": 5051.899999976158,
      "megapixelsPerSecond": 0.8103369397217929,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4728.399999976158,
      "inferenceMs": 172.19999998807907,
      "totalMs": 4900.599999964237,
      "megapixelsPerSecond": 1.5223228804770468,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4363.5,
      "inferenceMs": 872.3000000119209,
      "totalMs": 5235.800000011921,
      "megapixelsPerSecond": 0.801779204391198,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4363.5,
      "inferenceMs": 635.9000000059605,
      "totalMs": 4999.4000000059605,
      "megapixelsPerSecond": 0.41224091837952953,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4363.5,
      "inferenceMs": 535.7000000178814,
      "totalMs": 4899.200000017881,
      "megapixelsPerSecond": 0.48934851594409134,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4563.699999988079,
      "inferenceMs": 607.5,
      "totalMs": 5171.199999988079,
      "megapixelsPerSecond": 1.1512625514403292,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4563.699999988079,
      "inferenceMs": 294.90000000596046,
      "totalMs": 4858.5999999940395,
      "megapixelsPerSecond": 0.8889250593241831,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4563.699999988079,
      "inferenceMs": 157.19999998807907,
      "totalMs": 4720.899999976158,
      "megapixelsPerSecond": 1.6675826973274754,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4516.5999999940395,
      "inferenceMs": 997.9000000059605,
      "totalMs": 5514.5,
      "megapixelsPerSecond": 0.7008638140052336,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4516.5999999940395,
      "inferenceMs": 678,
      "totalMs": 5194.5999999940395,
      "megapixelsPerSecond": 0.3866430678466076,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4516.5999999940395,
      "inferenceMs": 544.5999999940395,
      "totalMs": 5061.199999988079,
      "megapixelsPerSecond": 0.48135145061121754,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4917.799999982119,
      "inferenceMs": 908.6999999880791,
      "totalMs": 5826.499999970198,
      "megapixelsPerSecond": 0.7696621547366294,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4917.799999982119,
      "inferenceMs": 659.2000000178814,
      "totalMs": 5577,
      "megapixelsPerSecond": 0.3976699029018342,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4917.799999982119,
      "inferenceMs": 553.2000000178814,
      "totalMs": 5471,
      "megapixelsPerSecond": 0.4738684020092671,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "64e537ca2bf6bf2afc13727ee6bb61cfca9e1048",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B0 · Cityscapes (640×1280)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4762.5999999940395,
      "inferenceMs": 693.0999999940395,
      "totalMs": 5455.699999988079,
      "megapixelsPerSecond": 1.00907805512338,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4762.5999999940395,
      "inferenceMs": 345.5,
      "totalMs": 5108.0999999940395,
      "megapixelsPerSecond": 0.7587380607814761,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4762.5999999940395,
      "inferenceMs": 160.90000000596046,
      "totalMs": 4923.5,
      "megapixelsPerSecond": 1.6292355499707207,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4556.800000011921,
      "inferenceMs": 985.2000000178814,
      "totalMs": 5542.000000029802,
      "megapixelsPerSecond": 0.7098984977540662,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4556.800000011921,
      "inferenceMs": 659.2999999821186,
      "totalMs": 5216.0999999940395,
      "megapixelsPerSecond": 0.3976095859352492,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4556.800000011921,
      "inferenceMs": 520.5999999940395,
      "totalMs": 5077.4000000059605,
      "megapixelsPerSecond": 0.5035420668517122,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5145.5,
      "inferenceMs": 645.6000000238419,
      "totalMs": 5791.100000023842,
      "megapixelsPerSecond": 1.0833209417195966,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5145.5,
      "inferenceMs": 326.40000000596046,
      "totalMs": 5471.9000000059605,
      "megapixelsPerSecond": 0.8031372548872945,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5145.5,
      "inferenceMs": 156,
      "totalMs": 5301.5,
      "megapixelsPerSecond": 1.6804102564102563,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4708.800000011921,
      "inferenceMs": 1141.9000000059605,
      "totalMs": 5850.700000017881,
      "megapixelsPerSecond": 0.6124809527947713,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4708.800000011921,
      "inferenceMs": 678.9000000059605,
      "totalMs": 5387.700000017881,
      "megapixelsPerSecond": 0.3861305052256569,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4708.800000011921,
      "inferenceMs": 566.8000000119209,
      "totalMs": 5275.600000023842,
      "megapixelsPerSecond": 0.46249823569951765,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4922,
      "inferenceMs": 1134.800000011921,
      "totalMs": 6056.800000011921,
      "megapixelsPerSecond": 0.6163130066907411,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4922,
      "inferenceMs": 685.5999999940395,
      "totalMs": 5607.5999999940395,
      "megapixelsPerSecond": 0.38235705951324245,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4922,
      "inferenceMs": 580.3000000119209,
      "totalMs": 5502.300000011921,
      "megapixelsPerSecond": 0.4517387558066773,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "ae0e7d35a49bb915f33a58b2ea93f4a6323bcb3f",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B0 · Cityscapes (768×768)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
afb669d7cfdf8ea51954e52513fc5660a52f1a06
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4904.799999982119,
      "inferenceMs": 569.0999999940395,
      "totalMs": 5473.899999976158,
      "megapixelsPerSecond": 1.2289439465951943,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4904.799999982119,
      "inferenceMs": 264,
      "totalMs": 5168.799999982119,
      "megapixelsPerSecond": 0.9929696969696968,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4904.799999982119,
      "inferenceMs": 142.09999999403954,
      "totalMs": 5046.899999976158,
      "megapixelsPerSecond": 1.8447853624982111,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4452.9000000059605,
      "inferenceMs": 930.3000000119209,
      "totalMs": 5383.200000017881,
      "megapixelsPerSecond": 0.7517918950779726,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4452.9000000059605,
      "inferenceMs": 650.4000000059605,
      "totalMs": 5103.300000011921,
      "megapixelsPerSecond": 0.40305043050061135,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4452.9000000059605,
      "inferenceMs": 526,
      "totalMs": 4978.9000000059605,
      "megapixelsPerSecond": 0.49837262357414447,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4589,
      "inferenceMs": 596.0999999940395,
      "totalMs": 5185.0999999940395,
      "megapixelsPerSecond": 1.1732796510769894,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4589,
      "inferenceMs": 306.40000000596046,
      "totalMs": 4895.4000000059605,
      "megapixelsPerSecond": 0.8555613576857064,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4589,
      "inferenceMs": 158.90000000596046,
      "totalMs": 4747.9000000059605,
      "megapixelsPerSecond": 1.649741976023705,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5113.0999999940395,
      "inferenceMs": 1087.9000000059605,
      "totalMs": 6201,
      "megapixelsPerSecond": 0.6428826178841512,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5113.0999999940395,
      "inferenceMs": 658.7000000178814,
      "totalMs": 5771.800000011921,
      "megapixelsPerSecond": 0.39797176255181976,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5113.0999999940395,
      "inferenceMs": 555.5,
      "totalMs": 5668.5999999940395,
      "megapixelsPerSecond": 0.47190639063906387,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4366,
      "inferenceMs": 961.2999999821186,
      "totalMs": 5327.299999982119,
      "megapixelsPerSecond": 0.7275481119452925,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4366,
      "inferenceMs": 660.5999999940395,
      "totalMs": 5026.5999999940395,
      "megapixelsPerSecond": 0.3968271268579553,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4366,
      "inferenceMs": 548.5,
      "totalMs": 4914.5,
      "megapixelsPerSecond": 0.4779288969917958,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "afb669d7cfdf8ea51954e52513fc5660a52f1a06",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B1 · ADE20K (512×512)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
c55f9dc9b8d8074a001fbdefc33c0d720d648bee
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5260.699999988079,
      "inferenceMs": 1139.7000000178814,
      "totalMs": 6400.4000000059605,
      "megapixelsPerSecond": 0.613663244703893,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5260.699999988079,
      "inferenceMs": 533.7000000178814,
      "totalMs": 5794.4000000059605,
      "megapixelsPerSecond": 0.4911823121439328,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5260.699999988079,
      "inferenceMs": 367.10000002384186,
      "totalMs": 5627.800000011921,
      "megapixelsPerSecond": 0.7140942522009661,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5390,
      "inferenceMs": 1685.2999999821186,
      "totalMs": 7075.299999982119,
      "megapixelsPerSecond": 0.4149955497581563,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5390,
      "inferenceMs": 1209.4000000059605,
      "totalMs": 6599.4000000059605,
      "megapixelsPerSecond": 0.2167554159076468,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5390,
      "inferenceMs": 1102.9000000059605,
      "totalMs": 6492.9000000059605,
      "megapixelsPerSecond": 0.2376861002797926,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4505,
      "inferenceMs": 1098.699999988079,
      "totalMs": 5603.699999988079,
      "megapixelsPerSecond": 0.6365632110745321,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4505,
      "inferenceMs": 528.5999999940395,
      "totalMs": 5033.5999999940395,
      "megapixelsPerSecond": 0.4959213015568595,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4505,
      "inferenceMs": 366.09999999403954,
      "totalMs": 4871.0999999940395,
      "megapixelsPerSecond": 0.7160447965153454,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4740.4000000059605,
      "inferenceMs": 1819.5999999940395,
      "totalMs": 6560,
      "megapixelsPerSecond": 0.3843657946814086,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4740.4000000059605,
      "inferenceMs": 1261.2999999821186,
      "totalMs": 6001.699999988079,
      "megapixelsPerSecond": 0.20783635931476763,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4740.4000000059605,
      "inferenceMs": 1109.7000000178814,
      "totalMs": 5850.100000023842,
      "megapixelsPerSecond": 0.23622961160293401,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4900.800000011921,
      "inferenceMs": 1733.199999988079,
      "totalMs": 6634,
      "megapixelsPerSecond": 0.4035264251123993,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4900.800000011921,
      "inferenceMs": 1277,
      "totalMs": 6177.800000011921,
      "megapixelsPerSecond": 0.2052811276429131,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4900.800000011921,
      "inferenceMs": 1119.0999999940395,
      "totalMs": 6019.9000000059605,
      "megapixelsPerSecond": 0.23424537574961682,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "c55f9dc9b8d8074a001fbdefc33c0d720d648bee",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B1 · Cityscapes (1024×1024)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
0dbec5d9127b402e2adba8aa13a380ff7e52e06a
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5995.699999988079,
      "inferenceMs": 480.90000000596046,
      "totalMs": 6476.5999999940395,
      "megapixelsPerSecond": 1.4543397795619286,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5995.699999988079,
      "inferenceMs": 418.2000000178814,
      "totalMs": 6413.9000000059605,
      "megapixelsPerSecond": 0.626838833067411,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5995.699999988079,
      "inferenceMs": 361,
      "totalMs": 6356.699999988079,
      "megapixelsPerSecond": 0.7261606648199446,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4774.800000011921,
      "inferenceMs": 4618.5,
      "totalMs": 9393.300000011921,
      "megapixelsPerSecond": 0.15143271624986468,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4774.800000011921,
      "inferenceMs": 4572,
      "totalMs": 9346.800000011921,
      "megapixelsPerSecond": 0.057336832895888014,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4774.800000011921,
      "inferenceMs": 4454.800000011921,
      "totalMs": 9229.600000023842,
      "megapixelsPerSecond": 0.058845290473039984,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4585.199999988079,
      "inferenceMs": 713.1999999880791,
      "totalMs": 5298.399999976158,
      "megapixelsPerSecond": 0.9806393718615959,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4585.199999988079,
      "inferenceMs": 415.59999999403954,
      "totalMs": 5000.799999982119,
      "megapixelsPerSecond": 0.630760346496053,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4585.199999988079,
      "inferenceMs": 295.5,
      "totalMs": 4880.699999988079,
      "megapixelsPerSecond": 0.8871201353637902,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4806.800000011921,
      "inferenceMs": 3461.600000023842,
      "totalMs": 8268.400000035763,
      "megapixelsPerSecond": 0.20204298590108127,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4806.800000011921,
      "inferenceMs": 3216.800000011921,
      "totalMs": 8023.600000023842,
      "megapixelsPerSecond": 0.0814921661275269,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4806.800000011921,
      "inferenceMs": 3214.7999999821186,
      "totalMs": 8021.5999999940395,
      "megapixelsPerSecond": 0.0815428642532842,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7106,
      "inferenceMs": 4792.600000023842,
      "totalMs": 11898.600000023842,
      "megapixelsPerSecond": 0.14593164461806135,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7106,
      "inferenceMs": 4720.100000023842,
      "totalMs": 11826.100000023842,
      "megapixelsPerSecond": 0.055537806402126196,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7106,
      "inferenceMs": 4575.199999988079,
      "totalMs": 11681.199999988079,
      "megapixelsPerSecond": 0.057296730197736276,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "0dbec5d9127b402e2adba8aa13a380ff7e52e06a",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.0/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B2 · ADE20K (512×512)

16 configurations tested
Passed14
Failed2
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP16GPU · WebGPUFailed—
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
df795789e70f4089c8658907679c6fd2367c89a5
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4998.5999999940395,
      "inferenceMs": 1060.4000000059605,
      "totalMs": 6059,
      "megapixelsPerSecond": 0.6595548849453684,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.8571428571428571,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4998.5999999940395,
      "inferenceMs": 590,
      "totalMs": 5588.5999999940395,
      "megapixelsPerSecond": 0.44431186440677967,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4998.5999999940395,
      "inferenceMs": 399.2000000178814,
      "totalMs": 5397.800000011921,
      "megapixelsPerSecond": 0.6566733466639723,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4757.700000017881,
      "inferenceMs": 3592.0999999940395,
      "totalMs": 8349.800000011921,
      "megapixelsPerSecond": 0.1947028200777151,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.8571428571428571,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4757.700000017881,
      "inferenceMs": 3255.300000011921,
      "totalMs": 8013.000000029802,
      "megapixelsPerSecond": 0.08052836912083065,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4757.700000017881,
      "inferenceMs": 3200.7000000178814,
      "totalMs": 7958.400000035763,
      "megapixelsPerSecond": 0.08190208391868511,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5024.700000017881,
      "inferenceMs": 1019.7000000178814,
      "totalMs": 6044.400000035763,
      "megapixelsPerSecond": 0.6858801608195897,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.7142857142857143,
      "negativePassed": true,
      "passed": false
    },
    "runtime": {
      "loadMs": 5024.700000017881,
      "inferenceMs": 600.4000000059605,
      "totalMs": 5625.100000023842,
      "megapixelsPerSecond": 0.43661558960259417,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5024.700000017881,
      "inferenceMs": 383,
      "totalMs": 5407.700000017881,
      "megapixelsPerSecond": 0.6844490861618798,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5757.300000011921,
      "inferenceMs": 3724.300000011921,
      "totalMs": 9481.600000023842,
      "megapixelsPerSecond": 0.18779153129387036,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.8571428571428571,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5757.300000011921,
      "inferenceMs": 3363.0999999940395,
      "totalMs": 9120.40000000596,
      "megapixelsPerSecond": 0.07794713211039356,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5757.300000011921,
      "inferenceMs": 3200,
      "totalMs": 8957.300000011921,
      "megapixelsPerSecond": 0.08191999999999999,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4653.300000011921,
      "inferenceMs": 3698.4000000059605,
      "totalMs": 8351.700000017881,
      "megapixelsPerSecond": 0.18910664070919123,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.8571428571428571,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4653.300000011921,
      "inferenceMs": 3401.7000000178814,
      "totalMs": 8055.000000029802,
      "megapixelsPerSecond": 0.07706264514760915,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4653.300000011921,
      "inferenceMs": 3222.899999976158,
      "totalMs": 7876.199999988079,
      "megapixelsPerSecond": 0.08133792547145094,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "df795789e70f4089c8658907679c6fd2367c89a5",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B2 · Cityscapes (1024×1024)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
178592c2b7cd2025e0d5e8c32a5356f36fbde311
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 101684,
      "inferenceMs": 493.5,
      "totalMs": 102177.5,
      "megapixelsPerSecond": 1.4172077001013172,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 101684,
      "inferenceMs": 268.2000000178814,
      "totalMs": 101952.20000001788,
      "megapixelsPerSecond": 0.9774198358781594,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 101684,
      "inferenceMs": 191.2000000178814,
      "totalMs": 101875.20000001788,
      "megapixelsPerSecond": 1.3710460249763796,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4816.0999999940395,
      "inferenceMs": 3157.800000011921,
      "totalMs": 7973.9000000059605,
      "megapixelsPerSecond": 0.22148077775582992,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4816.0999999940395,
      "inferenceMs": 2928,
      "totalMs": 7744.0999999940395,
      "megapixelsPerSecond": 0.08953005464480875,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4816.0999999940395,
      "inferenceMs": 2999.4000000059605,
      "totalMs": 7815.5,
      "megapixelsPerSecond": 0.08739881309577884,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 10139.40000000596,
      "inferenceMs": 473.19999998807907,
      "totalMs": 10612.59999999404,
      "megapixelsPerSecond": 1.4780050718884599,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 10139.40000000596,
      "inferenceMs": 317.2999999821186,
      "totalMs": 10456.699999988079,
      "megapixelsPerSecond": 0.8261708163087711,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 10139.40000000596,
      "inferenceMs": 201.7000000178814,
      "totalMs": 10341.100000023842,
      "megapixelsPerSecond": 1.2996727812432327,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7346,
      "inferenceMs": 3261.2000000178814,
      "totalMs": 10607.200000017881,
      "megapixelsPerSecond": 0.21445848153936134,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7346,
      "inferenceMs": 2951.0999999940395,
      "totalMs": 10297.09999999404,
      "megapixelsPerSecond": 0.08882925011030783,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7346,
      "inferenceMs": 2979.899999976158,
      "totalMs": 10325.899999976158,
      "megapixelsPerSecond": 0.0879707372737667,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 90656.09999999404,
      "inferenceMs": 3337.2999999821186,
      "totalMs": 93993.39999997616,
      "megapixelsPerSecond": 0.20956821382667049,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 90656.09999999404,
      "inferenceMs": 3029.399999976158,
      "totalMs": 93685.4999999702,
      "megapixelsPerSecond": 0.08653330692614482,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 90656.09999999404,
      "inferenceMs": 3034.9000000059605,
      "totalMs": 93691,
      "megapixelsPerSecond": 0.08637648686924944,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "178592c2b7cd2025e0d5e8c32a5356f36fbde311",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:53988/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:53988/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:53988/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:53988/dist/index.mjs:687:42)"
  }
]

SegFormer B3 · Cityscapes (1024×1024)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
b039a1230a59a7bb18f4d222dcef0eb092c77d9e
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5557.100000023842,
      "inferenceMs": 380,
      "totalMs": 5937.100000023842,
      "megapixelsPerSecond": 1.8405052631578946,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5557.100000023842,
      "inferenceMs": 331.09999999403954,
      "totalMs": 5888.200000017881,
      "megapixelsPerSecond": 0.7917366354718185,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5557.100000023842,
      "inferenceMs": 249,
      "totalMs": 5806.100000023842,
      "megapixelsPerSecond": 1.0527871485943774,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5214.4000000059605,
      "inferenceMs": 3800.4000000059605,
      "totalMs": 9014.800000011921,
      "megapixelsPerSecond": 0.18403115461501504,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5214.4000000059605,
      "inferenceMs": 3722.2999999821186,
      "totalMs": 8936.699999988079,
      "megapixelsPerSecond": 0.07042527469609094,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5214.4000000059605,
      "inferenceMs": 3758.9000000059605,
      "totalMs": 8973.300000011921,
      "megapixelsPerSecond": 0.06973955146441362,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6027,
      "inferenceMs": 430.69999998807907,
      "totalMs": 6457.699999988079,
      "megapixelsPerSecond": 1.62384954729361,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6027,
      "inferenceMs": 333,
      "totalMs": 6360,
      "megapixelsPerSecond": 0.7872192192192191,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6027,
      "inferenceMs": 199.80000001192093,
      "totalMs": 6226.800000011921,
      "megapixelsPerSecond": 1.3120320319537504,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5592,
      "inferenceMs": 4155.699999988079,
      "totalMs": 9747.699999988079,
      "megapixelsPerSecond": 0.1682970378039816,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5592,
      "inferenceMs": 4083.9000000059605,
      "totalMs": 9675.90000000596,
      "megapixelsPerSecond": 0.06418962266451612,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5592,
      "inferenceMs": 3691.5999999940395,
      "totalMs": 9283.59999999404,
      "megapixelsPerSecond": 0.07101094376433613,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5568.4000000059605,
      "inferenceMs": 4075.5,
      "totalMs": 9643.90000000596,
      "megapixelsPerSecond": 0.17160888234572447,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5568.4000000059605,
      "inferenceMs": 3930.300000011921,
      "totalMs": 9498.700000017881,
      "megapixelsPerSecond": 0.06669821642093603,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5568.4000000059605,
      "inferenceMs": 3891.800000011921,
      "totalMs": 9460.200000017881,
      "megapixelsPerSecond": 0.067358034842283,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "b039a1230a59a7bb18f4d222dcef0eb092c77d9e",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/layer_norm.3/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64210/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64210/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64210/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64210/dist/index.mjs:687:42)"
  }
]

SegFormer B4 · ADE20K (512×512)

10 configurations tested
Passed6
Failed4
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMFailed—
Q8GPU · WebGPUFailed—
Q8CPU · WASMFailed—
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
0f92f0b465567a1a1f004ec33110992b6158edbc
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5829.5,
      "inferenceMs": 957.4000000059605,
      "totalMs": 6786.9000000059605,
      "megapixelsPerSecond": 0.7305118027947001,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5829.5,
      "inferenceMs": 681.5999999940395,
      "totalMs": 6511.0999999940395,
      "megapixelsPerSecond": 0.3846009389704994,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5829.5,
      "inferenceMs": 447.2000000178814,
      "totalMs": 6276.700000017881,
      "megapixelsPerSecond": 0.5861896243057202,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5293.5,
      "inferenceMs": 5114.5,
      "totalMs": 10408,
      "megapixelsPerSecond": 0.1367468960797732,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5355,
      "inferenceMs": 1292.2999999821186,
      "totalMs": 6647.299999982119,
      "megapixelsPerSecond": 0.5411994119087499,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5355,
      "inferenceMs": 830.5999999940395,
      "totalMs": 6185.5999999940395,
      "megapixelsPerSecond": 0.3156079942233098,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5355,
      "inferenceMs": 497.90000000596046,
      "totalMs": 5852.9000000059605,
      "megapixelsPerSecond": 0.526499297041297,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5382,
      "inferenceMs": 5384.9000000059605,
      "totalMs": 10766.90000000596,
      "megapixelsPerSecond": 0.1298802206167665,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 4957.5,
      "inferenceMs": 5408.899999976158,
      "totalMs": 10366.399999976158,
      "megapixelsPerSecond": 0.1293039250130494,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "0f92f0b465567a1a1f004ec33110992b6158edbc",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B4 · Cityscapes (1024×1024)

16 configurations tested
Passed15
Failed1
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
Q8GPU · WebGPUAcceleratedPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
FP32CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
c77f9ce627f573eacbbbb0745ab46ad7ce950fa1
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6042.700000017881,
      "inferenceMs": 565.2000000178814,
      "totalMs": 6607.900000035763,
      "megapixelsPerSecond": 1.237423920696874,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6042.700000017881,
      "inferenceMs": 429,
      "totalMs": 6471.700000017881,
      "megapixelsPerSecond": 0.611058275058275,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6042.700000017881,
      "inferenceMs": 231.2000000178814,
      "totalMs": 6273.900000035763,
      "megapixelsPerSecond": 1.133840830362134,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5168.5,
      "inferenceMs": 4574.699999988079,
      "totalMs": 9743.199999988079,
      "megapixelsPerSecond": 0.15288259339450075,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5168.5,
      "inferenceMs": 4716.5,
      "totalMs": 9885,
      "megapixelsPerSecond": 0.05558019718011237,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5168.5,
      "inferenceMs": 4564.5999999940395,
      "totalMs": 9733.09999999404,
      "megapixelsPerSecond": 0.05742978574252777,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5411.200000017881,
      "inferenceMs": 566,
      "totalMs": 5977.200000017881,
      "megapixelsPerSecond": 1.2356749116607775,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5411.200000017881,
      "inferenceMs": 353.40000000596046,
      "totalMs": 5764.600000023842,
      "megapixelsPerSecond": 0.7417770231906584,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5411.200000017881,
      "inferenceMs": 262.5,
      "totalMs": 5673.700000017881,
      "megapixelsPerSecond": 0.9986438095238095,
      "grade": "fast",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6328.5,
      "inferenceMs": 4811.5999999940395,
      "totalMs": 11140.09999999404,
      "megapixelsPerSecond": 0.145355391138263,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6328.5,
      "inferenceMs": 4561.9000000059605,
      "totalMs": 10890.40000000596,
      "megapixelsPerSecond": 0.057463776058146274,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6328.5,
      "inferenceMs": 4472,
      "totalMs": 10800.5,
      "megapixelsPerSecond": 0.058618962432915915,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5164.799999982119,
      "inferenceMs": 4845.600000023842,
      "totalMs": 10010.40000000596,
      "megapixelsPerSecond": 0.144335479609658,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5164.799999982119,
      "inferenceMs": 4868.4000000059605,
      "totalMs": 10033.199999988079,
      "megapixelsPerSecond": 0.0538460274422149,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5164.799999982119,
      "inferenceMs": 4759.4000000059605,
      "totalMs": 9924.199999988079,
      "megapixelsPerSecond": 0.05507921166526698,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 512,
        "width": 512
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "c77f9ce627f573eacbbbb0745ab46ad7ce950fa1",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64210/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64210/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64210/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64210/dist/index.mjs:687:42)"
  }
]

SegFormer B5 · ADE20K (640×640)

10 configurations tested
Passed6
Failed4
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMFailed—
Q8GPU · WebGPUFailed—
Q8CPU · WASMFailed—
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
02e76595880a97cf256ceecb93acbc070e21fe97
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 8413.100000023842,
      "inferenceMs": 1471.300000011921,
      "totalMs": 9884.400000035763,
      "megapixelsPerSecond": 0.47535648745621784,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 8413.100000023842,
      "inferenceMs": 794.4000000059605,
      "totalMs": 9207.500000029802,
      "megapixelsPerSecond": 0.32998992950406986,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 8413.100000023842,
      "inferenceMs": 572.6000000238419,
      "totalMs": 8985.700000047684,
      "megapixelsPerSecond": 0.45781348234209723,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 7309.4000000059605,
      "inferenceMs": 9202.700000017881,
      "totalMs": 16512.100000023842,
      "megapixelsPerSecond": 0.07599856563819761,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5545.4000000059605,
      "inferenceMs": 1078.5,
      "totalMs": 6623.9000000059605,
      "megapixelsPerSecond": 0.6484858599907278,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": false,
          "actualLabels": [
            "sidewalk"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 0.8571428571428571,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5545.4000000059605,
      "inferenceMs": 689.6999999880791,
      "totalMs": 6235.0999999940395,
      "megapixelsPerSecond": 0.38008409454042474,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5545.4000000059605,
      "inferenceMs": 497.59999999403954,
      "totalMs": 6043,
      "megapixelsPerSecond": 0.5268167202635451,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5787.0999999940395,
      "inferenceMs": 9245.5,
      "totalMs": 15032.59999999404,
      "megapixelsPerSecond": 0.07564674706614029,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5655.5,
      "inferenceMs": 9895.90000000596,
      "totalMs": 15551.40000000596,
      "megapixelsPerSecond": 0.07067492597940346,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 640,
        "width": 640
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "02e76595880a97cf256ceecb93acbc070e21fe97",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64063/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64063/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64063/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64063/dist/index.mjs:687:42)"
  }
]

SegFormer B5 · Cityscapes (1024×1024)

10 configurations tested
Passed5
Failed5
TestedNov 12, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUFailed—
Q8CPU · WASMFailed—
FP32CPU · WASMFailed—
FP16GPU · WebGPUFailed—
FP16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen interior-point classification and complete binary semantic partition; <=5000ms. See research/segformer-family-expansion/quality-gates.json.
Tested
2026-11-12T16:51:44.339Z
Environment
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/157.0.0.0 Safari/537.36; Apple M3 Max 96GiB; macOS27.2 26B118
Runtime
Transformers.js 4.2.0
Artifact revision
15093b37e2614e7dec7c5ef75fa475b096956b90
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg"
}
Rerun when
artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 6484.5,
      "inferenceMs": 20904.40000000596,
      "totalMs": 27388.90000000596,
      "megapixelsPerSecond": 0.033456688544029035,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5797.4000000059605,
      "inferenceMs": 27022.5,
      "totalMs": 32819.90000000596,
      "megapixelsPerSecond": 0.025881839208067352,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "q8",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "q8",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5795,
      "inferenceMs": 972.8000000119209,
      "totalMs": 6767.800000011921,
      "megapixelsPerSecond": 0.7189473684122425,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5795,
      "inferenceMs": 897.5999999940395,
      "totalMs": 6692.5999999940395,
      "megapixelsPerSecond": 0.2920499108753796,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5795,
      "inferenceMs": 817.0999999940395,
      "totalMs": 6612.0999999940395,
      "megapixelsPerSecond": 0.3208224207586737,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5685,
      "inferenceMs": 27299.5,
      "totalMs": 32984.5,
      "megapixelsPerSecond": 0.025619223795307608,
      "grade": "slow",
      "version": "4.2.0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "fallback": false,
      "gpuForwardQueueObserved": false,
      "acceleration": "cpu",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp32",
      "provider": "ONNX Runtime WebAssembly CPU",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "person",
          "xy": [
            450,
            400
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            500,
            300
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        },
        {
          "label": "person",
          "xy": [
            400,
            550
          ],
          "passed": true,
          "actualLabels": [
            "person"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5815.4000000059605,
      "inferenceMs": 888.4000000059605,
      "totalMs": 6703.800000011921,
      "megapixelsPerSecond": 0.7872489869375368,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "completeSemanticPartition": true,
      "points": [
        {
          "label": "car",
          "xy": [
            255,
            350
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            200,
            370
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "car",
          "xy": [
            330,
            330
          ],
          "passed": true,
          "actualLabels": [
            "car"
          ]
        },
        {
          "label": "road",
          "xy": [
            30,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "road",
          "xy": [
            470,
            475
          ],
          "passed": true,
          "actualLabels": [
            "road"
          ]
        },
        {
          "label": "building",
          "xy": [
            450,
            100
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        },
        {
          "label": "building",
          "xy": [
            240,
            60
          ],
          "passed": true,
          "actualLabels": [
            "building"
          ]
        }
      ],
      "pointAccuracy": 1,
      "negativePassed": true,
      "passed": true
    },
    "runtime": {
      "loadMs": 5815.4000000059605,
      "inferenceMs": 669.8999999761581,
      "totalMs": 6485.299999982119,
      "megapixelsPerSecond": 0.3913181071941032,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "completeSemanticPartition": true,
      "points": [],
      "pointAccuracy": 1,
      "negativePassed": false,
      "passed": false
    },
    "runtime": {
      "loadMs": 5815.4000000059605,
      "inferenceMs": 631.8000000119209,
      "totalMs": 6447.200000017881,
      "megapixelsPerSecond": 0.4149161126860617,
      "grade": "moderate",
      "version": "4.2.0",
      "actualDevice": "webgpu",
      "actualPrecision": "fp16",
      "fallback": false,
      "gpuForwardQueueObserved": true,
      "acceleration": "partial",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "loadTimingScope": "Pipeline/model initialization in a ready module worker with a local pinned model snapshot; runtime import and Internet transfer excluded.",
      "totalTimingScope": "Sum of measured model initialization and this fixture generation; excludes other fixture calls and worker boot.",
      "requestedDevice": "webgpu",
      "requestedPrecision": "fp16",
      "provider": "ONNX Runtime WebGPU (partial; hardware adapter probe retained)",
      "processorDeclaredSize": {
        "height": 1024,
        "width": 1024
      }
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "version": "4.2.0",
      "artifactRevision": "15093b37e2614e7dec7c5ef75fa475b096956b90",
      "requestedDevice": "wasm",
      "requestedPrecision": "fp16"
    },
    "failure": "Error: page.evaluate: Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 1, ERROR_MESSAGE: /mnt/vss/_work/1/s/onnxruntime/core/graph/graph_utils.cc:30 int onnxruntime::graph_utils::GetIndexFromName(const Node &, const std::string &, bool) itr != node_args.end() was false. Attempting to get index by a name which does not exist:InsertedPrecisionFreeCast_/segformer/encoder/block.3.1/layer_norm_2/Constant_output_0for node: /segformer/encoder/patch_embeddings.0/layer_norm/Mul/SimplifiedLayerNormFusion/\n\n    at createError (http://127.0.0.1:64210/dist/index.mjs:13:17)\n    at createWorkerError (http://127.0.0.1:64210/dist/index.mjs:29:10)\n    at WorkerManager.parseWorkerError (http://127.0.0.1:64210/dist/index.mjs:713:12)\n    at Worker.handleMessage (http://127.0.0.1:64210/dist/index.mjs:687:42)"
  }
]

Something not quite right?

Tell us what happened with SegFormer. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.