{ "schema": "rtmw3d-head-experiments/v1", "sourceCommit": "40a8a49d99203b6c3afd654883c6c1696d6c3192", "keypointCount": 133, "variants": [ { "id": "rtmw3d-l-fp32-gpu-decoder", "model": "rtmw3d-l", "policy": "fp32-gpu-decoder", "file": "experiments/rtmw3d-l-fp32-gpu-decoder.onnx", "sizeBytes": 230006412, "sha256": "8fc094f02434d5de23cdc41755c3d88e8f46ed96bbb02cc0c7a80c04433977aa", "precision": "fp32", "experimental": true, "qualityPassed": true, "browserGPUValidated": false, "sourceRevision": "88ee5cb728cb2ed3efaf43011c9dd4908a6504b7", "sourceModelSha256": "4e37e5e9043817c5ff387b61c44bb48af94ad42e4978c4e43780761c65582abc", "upstream": "https://github.com/open-mmlab/mmpose", "weightsUnchangedFP32": true, "gpuDecoderMinimum": "Less+Where", "input": { "name": "input", "type": "float32", "shape": [ 1, 3, 384, 288 ] }, "outputs": [ { "name": "keypoints", "type": "float32", "shape": [ 1, 133, 3 ], "units": [ "crop-pixels", "crop-pixels", "root-relative-metres" ] }, { "name": "scores", "type": "float32", "shape": [ 1, 133 ] } ], "outputBytesPerFrame": 2128, "changedConvolutions": [], "comparisons": [ { "publicFixtureImageId": 40083, "parts": { "all": { "joints": 111, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 16, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 22, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 21, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 0.0 }, { "publicFixtureImageId": 197388, "parts": { "all": { "joints": 133, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 17, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 23, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 42, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 0.0 } ], "referenceLatencyMs": [], "note": "Same trained weights and 133-point 3D contract; explicit im2col and optional batched split-K reduction. CPU conversion tests are not WebGPU performance or target-video accuracy measurements." }, { "id": "rtmw3d-l-fp32-head-gemm-split8", "model": "rtmw3d-l", "policy": "fp32-head-gemm-split8", "file": "experiments/rtmw3d-l-fp32-head-gemm-split8.onnx", "sizeBytes": 230139583, "sha256": "5cde82faa4f8e626ad7594f1231508e7622a020d821038f02f962801cdda6819", "precision": "fp32", "experimental": true, "qualityPassed": true, "browserGPUValidated": false, "sourceRevision": "88ee5cb728cb2ed3efaf43011c9dd4908a6504b7", "sourceModelSha256": "4e37e5e9043817c5ff387b61c44bb48af94ad42e4978c4e43780761c65582abc", "upstream": "https://github.com/open-mmlab/mmpose", "weightsUnchangedFP32": true, "gpuDecoderMinimum": "Less+Where", "input": { "name": "input", "type": "float32", "shape": [ 1, 3, 384, 288 ] }, "outputs": [ { "name": "keypoints", "type": "float32", "shape": [ 1, 133, 3 ], "units": [ "crop-pixels", "crop-pixels", "root-relative-metres" ] }, { "name": "scores", "type": "float32", "shape": [ 1, 133 ] } ], "outputBytesPerFrame": 2128, "changedConvolutions": [ { "node": "/head/final_layer/conv/Conv", "inputShape": [ 1, 1024, 12, 9 ], "weightShape": [ 133, 1024, 7, 7 ], "outputShape": [ 1, 133, 12, 9 ], "kernel": [ 7, 7 ], "pads": [ 3, 3, 3, 3 ], "strides": [ 1, 1 ], "dilations": [ 1, 1 ], "splits": 8, "convolutionFLOPs": 1441456128, "columnBytes": 21676032 }, { "node": "/head/final_layer2/conv/Conv", "inputShape": [ 1, 768, 24, 18 ], "weightShape": [ 133, 768, 7, 7 ], "outputShape": [ 1, 133, 24, 18 ], "kernel": [ 7, 7 ], "pads": [ 3, 3, 3, 3 ], "strides": [ 1, 1 ], "dilations": [ 1, 1 ], "splits": 8, "convolutionFLOPs": 4324368384, "columnBytes": 65028096 } ], "comparisons": [ { "publicFixtureImageId": 40083, "parts": { "all": { "joints": 111, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 16, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 22, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 21, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 7.152557373046875e-07 }, { "publicFixtureImageId": 197388, "parts": { "all": { "joints": 133, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 17, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 23, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 42, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 4.76837158203125e-07 } ], "referenceLatencyMs": [], "note": "Same trained weights and 133-point 3D contract; explicit im2col and optional batched split-K reduction. CPU conversion tests are not WebGPU performance or target-video accuracy measurements." }, { "id": "rtmw3d-l-fp32-head-gemm", "model": "rtmw3d-l", "policy": "fp32-head-gemm", "file": "experiments/rtmw3d-l-fp32-head-gemm.onnx", "sizeBytes": 230137892, "sha256": "b09da548132bce0d04d99008e46b13f2bb3f4407a8513c5fa8f24a5b6d8071d8", "precision": "fp32", "experimental": true, "qualityPassed": true, "browserGPUValidated": false, "sourceRevision": "88ee5cb728cb2ed3efaf43011c9dd4908a6504b7", "sourceModelSha256": "4e37e5e9043817c5ff387b61c44bb48af94ad42e4978c4e43780761c65582abc", "upstream": "https://github.com/open-mmlab/mmpose", "weightsUnchangedFP32": true, "gpuDecoderMinimum": "Less+Where", "input": { "name": "input", "type": "float32", "shape": [ 1, 3, 384, 288 ] }, "outputs": [ { "name": "keypoints", "type": "float32", "shape": [ 1, 133, 3 ], "units": [ "crop-pixels", "crop-pixels", "root-relative-metres" ] }, { "name": "scores", "type": "float32", "shape": [ 1, 133 ] } ], "outputBytesPerFrame": 2128, "changedConvolutions": [ { "node": "/head/final_layer/conv/Conv", "inputShape": [ 1, 1024, 12, 9 ], "weightShape": [ 133, 1024, 7, 7 ], "outputShape": [ 1, 133, 12, 9 ], "kernel": [ 7, 7 ], "pads": [ 3, 3, 3, 3 ], "strides": [ 1, 1 ], "dilations": [ 1, 1 ], "splits": 1, "convolutionFLOPs": 1441456128, "columnBytes": 21676032 }, { "node": "/head/final_layer2/conv/Conv", "inputShape": [ 1, 768, 24, 18 ], "weightShape": [ 133, 768, 7, 7 ], "outputShape": [ 1, 133, 24, 18 ], "kernel": [ 7, 7 ], "pads": [ 3, 3, 3, 3 ], "strides": [ 1, 1 ], "dilations": [ 1, 1 ], "splits": 1, "convolutionFLOPs": 4324368384, "columnBytes": 65028096 } ], "comparisons": [ { "publicFixtureImageId": 40083, "parts": { "all": { "joints": 111, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 16, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 22, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 21, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 4.76837158203125e-07 }, { "publicFixtureImageId": 197388, "parts": { "all": { "joints": 133, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body": { "joints": 17, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "body-feet": { "joints": 23, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hips": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "knees": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "ankles": { "joints": 2, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "feet": { "joints": 6, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "face": { "joints": 68, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 }, "hands": { "joints": 42, "xyP95_px": 0.0, "zP95_m": 0.0, "xyMax_px": 0.0, "zMax_m": 0.0, "confidenceLossCount": 0 } }, "scoreMaxAbsoluteError": 4.76837158203125e-07 } ], "referenceLatencyMs": [], "note": "Same trained weights and 133-point 3D contract; explicit im2col and optional batched split-K reduction. CPU conversion tests are not WebGPU performance or target-video accuracy measurements." } ], "browserGPUValidated": false }