{
  "schema": "flabs-task-comparison/1",
  "title": "商品背景提取实测：U²-Net 与 U²-NetP 的 CPU 对比",
  "publisher": "FLABS 时尚AI实验室",
  "task": "foreground",
  "measuredAt": "2026-10-10T07:55:40.293075+00:00",
  "measurementKind": "actual-local-inference",
  "protocol": {
    "id": "foreground",
    "name": "商品背景提取",
    "backends": [
      "u2net",
      "u2netp"
    ],
    "runtime": "ONNX Runtime CPU, one inference thread; official rembg preprocessing and first output; no model-specific tuning",
    "inputs": "three private existing product photos; 600px long side, Lanczos",
    "qualityPolicy": "No manually labelled masks, so no IoU or quality score. Save alpha, outputs, diagnostic coverage and AI visual observations; human approval remains pending."
  },
  "protocolSHA256": "e5334784290c930b95b20128cf62b0e19bff0533c4c42be3503cf96d63955923",
  "runnerSHA256": "a6cf8966acf8a177e9f44f8f1f0960aab6e14a0251090aad5936a0aedef2f977",
  "environment": {
    "python": "3.12.13",
    "os": "Windows-11-10.0.22621-SP0",
    "processor": "AMD64 Family 25 Model 33 Stepping 0, AuthenticAMD",
    "runtimePackages": {
      "onnxruntime": "1.23.2",
      "opencv-contrib-python-headless": "4.12.0.88",
      "numpy": "2.2.6",
      "Pillow": "11.3.0"
    },
    "compute": "CPU only",
    "threads": 1,
    "onnxProviders": [
      "CPUExecutionProvider"
    ],
    "opencvOpenCL": false
  },
  "warmups": 6,
  "timedInferences": 18,
  "paidAPIRequests": 0,
  "priceNote": "仅记录本次本地调用无付费API；未计硬件、电费或商业服务价格。",
  "assetPermission": "用户授权的项目素材，仅用于私有测试；不提供公共下载。",
  "qualityScore": null,
  "humanAcceptance": false,
  "records": [
    {
      "model": "u2net",
      "inferenceMs": [
        1058.885,
        1034.186,
        1032.694
      ],
      "medianMs": 1034.186,
      "width": 450,
      "height": 600,
      "deterministicAcross3": true,
      "sample": "A",
      "sampleType": "衬衫",
      "inputSHA256": "7c901531bb6c27bf54af05cbed74523b228bc9f439eede6243e7086d3f5559dd",
      "outputSHA256": "2ca7e19878ff3b02d5fb338c702d1e45a85a5e73bfd73aa217162c83e7aaff2e",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    },
    {
      "model": "u2net",
      "inferenceMs": [
        1059.937,
        1058.456,
        1055.873
      ],
      "medianMs": 1058.456,
      "width": 600,
      "height": 448,
      "deterministicAcross3": true,
      "sample": "B",
      "sampleType": "长裤",
      "inputSHA256": "ad78f3db76c5b3c307a2fee22e2a174a70c57532090742412fb3926e735b5222",
      "outputSHA256": "dd5f043bdafcf48f3d09add8fea895b9aadd89b19e305fb128f7d77f0b51f04c",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    },
    {
      "model": "u2net",
      "inferenceMs": [
        1040.301,
        1043.551,
        1062.134
      ],
      "medianMs": 1043.551,
      "width": 384,
      "height": 600,
      "deterministicAcross3": true,
      "sample": "C",
      "sampleType": "鞋",
      "inputSHA256": "adca79e18859de767a9ac7e1846026070897b0177b813811c81cbc2cca0d98ec",
      "outputSHA256": "b48423f6d519bab5eb3eb73026441982a1c6c705d1d602afb414747797753d8c",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    },
    {
      "model": "u2netp",
      "inferenceMs": [
        446.819,
        432.659,
        421.469
      ],
      "medianMs": 432.659,
      "width": 450,
      "height": 600,
      "deterministicAcross3": true,
      "sample": "A",
      "sampleType": "衬衫",
      "inputSHA256": "7c901531bb6c27bf54af05cbed74523b228bc9f439eede6243e7086d3f5559dd",
      "outputSHA256": "99395141890f844f51eddb275681cf28e1d5779f28c14669469ebce65bf2cdc8",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    },
    {
      "model": "u2netp",
      "inferenceMs": [
        422.233,
        419.31,
        424.229
      ],
      "medianMs": 422.233,
      "width": 600,
      "height": 448,
      "deterministicAcross3": true,
      "sample": "B",
      "sampleType": "长裤",
      "inputSHA256": "ad78f3db76c5b3c307a2fee22e2a174a70c57532090742412fb3926e735b5222",
      "outputSHA256": "d9a81fd0c9041751bf3a44aa1fe91438cf3b16764dba62fec0193cd7346f7b29",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    },
    {
      "model": "u2netp",
      "inferenceMs": [
        429.448,
        422.028,
        419.786
      ],
      "medianMs": 422.028,
      "width": 384,
      "height": 600,
      "deterministicAcross3": true,
      "sample": "C",
      "sampleType": "鞋",
      "inputSHA256": "adca79e18859de767a9ac7e1846026070897b0177b813811c81cbc2cca0d98ec",
      "outputSHA256": "837c471e267e63cd8378d8c3f5dd7f5370d58cdd9f23f774eeac4ea4ec7f1b33",
      "maskAccuracy": null,
      "groundTruthMaskAvailable": false
    }
  ],
  "weights": [
    {
      "file": "u2net.onnx",
      "bytes": 175997641,
      "sha256": "8d10d2f3bb75ae3b6d527c77944fc5e7dcd94b29809d47a739a7a728a912b491",
      "downloadUrl": "https://github.com/danielgatis/rembg/releases/download/v0.0.0/u2net.onnx"
    },
    {
      "file": "u2netp.onnx",
      "bytes": 4574861,
      "sha256": "309c8469258dda742793dce0ebea8e6dd393174f89934733ecc8b14c76f4ddd8",
      "downloadUrl": "https://github.com/danielgatis/rembg/releases/download/v0.0.0/u2netp.onnx"
    }
  ],
  "observations": [
    "U²-NetP 在本机三个样本的计时均更短；各样本中位数的中位数为422.233ms，U²-Net为1043.551ms。",
    "AI查看私有输出后发现：两者保留衣架，鞋子样本仍有细背景线条；不能直接视为电商成品。",
    "没有人工标注的真值蒙版，未测IoU或分割准确率。"
  ],
  "limitations": [
    "仅3张商品照片、每张3次计时，不代表所有设计及生产任务。",
    "本地CPU单线程，未测试GPU、网络接口及商业平台。",
    "测试图片和输出仅在私有证据包留存；网站不分发商品原图。",
    "未取得用户视觉验收，不设置综合质量分或通用排名。"
  ]
}