fix(接口4): worker 移除脸型 Mock + face_shape 改本机 MediaPipe 计算

- worker /api/v1/face/features 不再返回假成功数据，直接告知仅网关实现，避免本机误打 :8187 被 Mock 结果误导。 - 网关 ark_api_key 加载优先级改为 gateway/config.json 优先（原先误读 worker_config.json 里的失效 key）。 - 接口4 face_shape 不再采信豆包结果，改用本机 face/face_shape_classifier.py （MediaPipe 7 类）计算覆盖；其余 5 项特征仍走豆包。 - 修复 face_shape_classifier 共享 FaceMesh 实例的线程安全问题（加锁），避免网关侧接口4 并发请求时崩溃/结果错乱。 - 新增 /api/v1/debug/face-shape 调试接口 + static/test_face_shape.html 单图调试页（worker 侧）。 - 更新文档：网关机现在也需要 mediapipe/opencv-python/numpy<2。 ⚠️ 部署前提醒：网关机需先安装 mediapipe==0.10.14 / opencv-python==4.10.0.84 / numpy==1.26.4，否则接口4 会返回 1007「分析服务异常」。 Co-authored-by: Cursor <cursoragent@cursor.com>
Merge branch 'main' of http://git.xiangsilian.com:3000/xsl/hair
2026-07-29 14:41:49 +08:00 · 2026-07-29 13:26:44 +08:00 · 2026-07-29 13:16:43 +08:00 · 2026-07-27 23:21:01 +08:00 · 2026-07-24 00:52:14 +08:00 · 2026-07-24 00:52:14 +08:00
335 changed files with 14245 additions and 498 deletions
@@ -53,3 +53,32 @@ static/report_hairline_v2.zip
 # local_test 运行期日志 / pid（不入 git）
 local_test/hair_service.log
 local_test/hair_service.pid
+
+# benchmark 原始输出（含结果图+原图，体积大，不入 git）
+benchmark_out/
+
+# benchmark 部署的 HTML 报告（图片 base64 内嵌，体积大，不入 git）
+static/hairstyle_thumbs/
+
+# 网关运行期日志（不入 git）
+gateway.log
+
+# 工作流备份文件（不入 git）
+*.json.bak.*
+
+# 脸型测试素材（人像照片，体积大，不入 git；仅保留 6 张基准标注图）
+face/test_img/脸型测试集合/
+face/test_img/girl/
+face/test_img/man/
+
+# 脸型特征缓存（由 face/dump_features.py 生成，可随时重跑）
+face/cache/
+
+# 脸型报告输出（标注图体积大，不入 git）
+static/face_shape_report/
+static/face_shape_report.html
+static/facetest_report/
+static/facetest_report.html
+static/facetest_all_report/
+static/facetest_all_report.html
+
@@ -0,0 +1,327 @@
+{
+  "16": {
+    "class_type": "UNETLoader",
+    "inputs": {
+      "unet_name": "flux-2-klein-4b-fp8.safetensors",
+      "weight_dtype": "fp8_e4m3fn_fast"
+    }
+  },
+  "3": {
+    "class_type": "VAELoader",
+    "inputs": {
+      "vae_name": "flux2-vae.safetensors"
+    }
+  },
+  "61": {
+    "class_type": "CLIPLoader",
+    "inputs": {
+      "clip_name": "qwen_3_4b.safetensors",
+      "type": "flux2",
+      "device": "cpu"
+    }
+  },
+  "26": {
+    "class_type": "LoadImage",
+    "inputs": {
+      "image": "placeholder.png"
+    }
+  },
+  "60": {
+    "class_type": "JjkText",
+    "inputs": {
+      "text": "填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜"
+    }
+  },
+  "22": {
+    "class_type": "CLIPTextEncode",
+    "inputs": {
+      "clip": [
+        "61",
+        0
+      ],
+      "text": [
+        "60",
+        0
+      ]
+    }
+  },
+  "31": {
+    "class_type": "easy imageSize",
+    "inputs": {
+      "image": [
+        "26",
+        0
+      ]
+    }
+  },
+  "33": {
+    "class_type": "Mask Fill Holes",
+    "inputs": {
+      "masks": [
+        "26",
+        1
+      ]
+    }
+  },
+  "36": {
+    "class_type": "Convert Masks to Images",
+    "inputs": {
+      "masks": [
+        "33",
+        0
+      ]
+    }
+  },
+  "39": {
+    "class_type": "ImageScale",
+    "inputs": {
+      "image": [
+        "36",
+        0
+      ],
+      "upscale_method": "nearest-exact",
+      "width": [
+        "31",
+        0
+      ],
+      "height": [
+        "31",
+        1
+      ],
+      "crop": "disabled"
+    }
+  },
+  "37": {
+    "class_type": "Image To Mask",
+    "inputs": {
+      "image": [
+        "39",
+        0
+      ],
+      "method": "intensity"
+    }
+  },
+  "32": {
+    "class_type": "LayerUtility: ImageScaleByAspectRatio V2",
+    "inputs": {
+      "image": [
+        "26",
+        0
+      ],
+      "mask": [
+        "37",
+        0
+      ],
+      "aspect_ratio": "custom",
+      "proportional_width": [
+        "31",
+        0
+      ],
+      "proportional_height": [
+        "31",
+        1
+      ],
+      "fit": "letterbox",
+      "method": "lanczos",
+      "round_to_multiple": "8",
+      "scale_to_side": "None",
+      "scale_to_length": 1024,
+      "background_color": "#000000"
+    }
+  },
+  "44": {
+    "class_type": "ImageAndMaskPreview",
+    "inputs": {
+      "image": [
+        "32",
+        0
+      ],
+      "mask": [
+        "32",
+        1
+      ],
+      "mask_opacity": 1,
+      "mask_color": "FFFF00",
+      "pass_through": true
+    }
+  },
+  "14": {
+    "class_type": "GetImageSize+",
+    "inputs": {
+      "image": [
+        "44",
+        0
+      ]
+    }
+  },
+  "13": {
+    "class_type": "VAEEncode",
+    "inputs": {
+      "pixels": [
+        "44",
+        0
+      ],
+      "vae": [
+        "3",
+        0
+      ]
+    }
+  },
+  "2": {
+    "class_type": "ModelSamplingFlux",
+    "inputs": {
+      "model": [
+        "16",
+        0
+      ],
+      "max_shift": 1.15,
+      "base_shift": 0.5,
+      "width": [
+        "14",
+        0
+      ],
+      "height": [
+        "14",
+        1
+      ]
+    }
+  },
+  "19": {
+    "class_type": "FluxGuidance",
+    "inputs": {
+      "conditioning": [
+        "22",
+        0
+      ],
+      "guidance": 1
+    }
+  },
+  "5": {
+    "class_type": "ReferenceLatent",
+    "inputs": {
+      "conditioning": [
+        "19",
+        0
+      ],
+      "latent": [
+        "13",
+        0
+      ]
+    }
+  },
+  "7": {
+    "class_type": "EmptySD3LatentImage",
+    "inputs": {
+      "width": [
+        "14",
+        0
+      ],
+      "height": [
+        "14",
+        1
+      ],
+      "batch_size": 1
+    }
+  },
+  "1": {
+    "class_type": "BasicScheduler",
+    "inputs": {
+      "model": [
+        "2",
+        0
+      ],
+      "scheduler": "simple",
+      "steps": 4,
+      "denoise": 1
+    }
+  },
+  "20": {
+    "class_type": "BasicGuider",
+    "inputs": {
+      "model": [
+        "2",
+        0
+      ],
+      "conditioning": [
+        "5",
+        0
+      ]
+    }
+  },
+  "6": {
+    "class_type": "RandomNoise",
+    "inputs": {
+      "noise_seed": 0
+    }
+  },
+  "8": {
+    "class_type": "KSamplerSelect",
+    "inputs": {
+      "sampler_name": "euler"
+    }
+  },
+  "9": {
+    "class_type": "SamplerCustomAdvanced",
+    "inputs": {
+      "noise": [
+        "6",
+        0
+      ],
+      "guider": [
+        "20",
+        0
+      ],
+      "sampler": [
+        "8",
+        0
+      ],
+      "sigmas": [
+        "1",
+        0
+      ],
+      "latent_image": [
+        "7",
+        0
+      ]
+    }
+  },
+  "10": {
+    "class_type": "VAEDecode",
+    "inputs": {
+      "samples": [
+        "9",
+        0
+      ],
+      "vae": [
+        "3",
+        0
+      ]
+    }
+  },
+  "62": {
+    "class_type": "ColorMatch",
+    "inputs": {
+      "image_ref": [
+        "26",
+        0
+      ],
+      "image_target": [
+        "10",
+        0
+      ],
+      "method": "mkl",
+      "strength": 1,
+      "multithread": true
+    }
+  },
+  "17": {
+    "class_type": "SaveImage",
+    "inputs": {
+      "images": [
+        "62",
+        0
+      ],
+      "filename_prefix": "hair_inpaint"
+    }
+  }
+}
@@ -0,0 +1,117 @@
+{
+  "16": {"class_type": "UNETLoader", "inputs": {
+    "unet_name": "flux2.0/flux-2-klein-9b-fp8.safetensors",
+    "weight_dtype": "fp8_e4m3fn_fast"}},
+  "3": {"class_type": "VAELoader", "inputs": {
+    "vae_name": "flux2-vae.safetensors"}},
+  "61": {"class_type": "CLIPLoader", "inputs": {
+    "clip_name": "qwen_3_8b_fp8mixed.safetensors",
+    "type": "flux2",
+    "device": "default"}},
+
+  "26": {"class_type": "LoadImage", "inputs": {
+    "image": "placeholder.png"}},
+
+  "60": {"class_type": "JjkText", "inputs": {
+    "text": "填充遮罩区域的头发，皮肤加一点磨皮"}},
+  "22": {"class_type": "CLIPTextEncode", "inputs": {
+    "clip": ["61", 0],
+    "text": ["60", 0]}},
+
+  "31": {"class_type": "easy imageSize", "inputs": {
+    "image": ["26", 0]}},
+
+  "33": {"class_type": "Mask Fill Holes", "inputs": {
+    "masks": ["26", 1]}},
+  "36": {"class_type": "Convert Masks to Images", "inputs": {
+    "masks": ["33", 0]}},
+  "39": {"class_type": "ImageScale", "inputs": {
+    "image": ["36", 0],
+    "upscale_method": "nearest-exact",
+    "width": ["31", 0],
+    "height": ["31", 1],
+    "crop": "disabled"}},
+  "37": {"class_type": "Image To Mask", "inputs": {
+    "image": ["39", 0],
+    "method": "intensity"}},
+
+  "32": {"class_type": "LayerUtility: ImageScaleByAspectRatio V2", "inputs": {
+    "image": ["26", 0],
+    "mask": ["37", 0],
+    "aspect_ratio": "custom",
+    "proportional_width": ["31", 0],
+    "proportional_height": ["31", 1],
+    "fit": "letterbox",
+    "method": "lanczos",
+    "round_to_multiple": "8",
+    "scale_to_side": "None",
+    "scale_to_length": 1024,
+    "background_color": "#000000"}},
+
+  "44": {"class_type": "ImageAndMaskPreview", "inputs": {
+    "image": ["32", 0],
+    "mask": ["32", 1],
+    "mask_opacity": 1,
+    "mask_color": "FFFF00",
+    "pass_through": true}},
+
+  "14": {"class_type": "GetImageSize+", "inputs": {
+    "image": ["44", 0]}},
+
+  "13": {"class_type": "VAEEncode", "inputs": {
+    "pixels": ["44", 0],
+    "vae": ["3", 0]}},
+
+  "2": {"class_type": "ModelSamplingFlux", "inputs": {
+    "model": ["16", 0],
+    "max_shift": 1.15,
+    "base_shift": 0.5,
+    "width": ["14", 0],
+    "height": ["14", 1]}},
+  "19": {"class_type": "FluxGuidance", "inputs": {
+    "conditioning": ["22", 0],
+    "guidance": 1}},
+  "5": {"class_type": "ReferenceLatent", "inputs": {
+    "conditioning": ["19", 0],
+    "latent": ["13", 0]}},
+
+  "7": {"class_type": "EmptySD3LatentImage", "inputs": {
+    "width": ["14", 0],
+    "height": ["14", 1],
+    "batch_size": 1}},
+
+  "1": {"class_type": "BasicScheduler", "inputs": {
+    "model": ["2", 0],
+    "scheduler": "simple",
+    "steps": 4,
+    "denoise": 1}},
+  "20": {"class_type": "BasicGuider", "inputs": {
+    "model": ["2", 0],
+    "conditioning": ["5", 0]}},
+
+  "6": {"class_type": "RandomNoise", "inputs": {
+    "noise_seed": 0}},
+  "8": {"class_type": "KSamplerSelect", "inputs": {
+    "sampler_name": "euler"}},
+  "9": {"class_type": "SamplerCustomAdvanced", "inputs": {
+    "noise": ["6", 0],
+    "guider": ["20", 0],
+    "sampler": ["8", 0],
+    "sigmas": ["1", 0],
+    "latent_image": ["7", 0]}},
+
+  "10": {"class_type": "VAEDecode", "inputs": {
+    "samples": ["9", 0],
+    "vae": ["3", 0]}},
+
+  "62": {"class_type": "ColorMatch", "inputs": {
+    "image_ref": ["26", 0],
+    "image_target": ["10", 0],
+    "method": "mkl",
+    "strength": 1,
+    "multithread": true}},
+
+  "17": {"class_type": "SaveImage", "inputs": {
+    "images": ["62", 0],
+    "filename_prefix": "hair_inpaint"}}
+}
@@ -82,7 +82,8 @@ model.safetensors         https://huggingface.co/jonathandinu/face-parsing/resol
 模型已就位，但**内网机还需要 Python 依赖的离线 wheel 包**，否则 `pip install` 在内网无法联网安装。这部分**与目标机的操作系统、Python 版本、CUDA 版本强相关**，需确认后单独打包：

 - **worker（GPU 机）**：`mediapipe` / `opencv-python` / `numpy<2` / `Pillow` / **`torch`+`torchvision` 的 CUDA 版**（按 GPU 的 CUDA 版本选 cu118/cu121 等）/ `transformers`（接口2 SegFormer）+ FastAPI/uvicorn 全家桶。
- **网关机**：很轻，只需 FastAPI/uvicorn/httpx 等代理依赖，**不需要 torch/mediapipe**。
+- **网关机**：FastAPI/uvicorn/httpx 等代理依赖 + **接口4 现需 `mediapipe`/`opencv-python`/`numpy<2`**（脸型本机计算），
+  仍**不需要 torch**（无 GPU 推理需求）。

 > 架构已拆分（见 `docs/实现说明.md`）：算法依赖只装在 worker，网关保持轻量。

@@ -2,7 +2,7 @@
  "1": {
    "inputs": {
      "scheduler": "simple",
-      "steps": 6,
+      "steps": 4,
      "denoise": 1,
      "model": [
        "2",
@@ -170,7 +170,7 @@
  },
  "16": {
    "inputs": {
-      "unet_name": "flux2.0/flux-2-klein-9b-fp8.safetensors",
+      "unet_name": "flux-2-klein-4b-fp8.safetensors",
      "weight_dtype": "fp8_e4m3fn"
    },
    "class_type": "UNETLoader",
@@ -410,7 +410,7 @@
  },
  "60": {
    "inputs": {
-      "text": "补充遮罩区补充遮罩区域内的头发，头发填满遮罩区域。发际线往下挡住额头"
+      "text": "填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜"
    },
    "class_type": "JjkText",
    "_meta": {
@@ -419,9 +419,9 @@
  },
  "61": {
    "inputs": {
-      "clip_name": "qwen_3_8b_fp8mixed.safetensors",
+      "clip_name": "qwen_3_4b.safetensors",
      "type": "flux2",
-      "device": "default"
+      "device": "cpu"
    },
    "class_type": "CLIPLoader",
    "_meta": {
@@ -447,4 +447,4 @@
      "title": "Color Match"
    }
  }
-}
+}
@@ -0,0 +1,450 @@
+{
+  "1": {
+    "inputs": {
+      "scheduler": "simple",
+      "steps": 6,
+      "denoise": 1,
+      "model": [
+        "2",
+        0
+      ]
+    },
+    "class_type": "BasicScheduler",
+    "_meta": {
+      "title": "基本调度器"
+    }
+  },
+  "2": {
+    "inputs": {
+      "max_shift": 1.15,
+      "base_shift": 0.5,
+      "width": [
+        "14",
+        0
+      ],
+      "height": [
+        "14",
+        1
+      ],
+      "model": [
+        "16",
+        0
+      ]
+    },
+    "class_type": "ModelSamplingFlux",
+    "_meta": {
+      "title": "采样算法（Flux）"
+    }
+  },
+  "3": {
+    "inputs": {
+      "vae_name": "flux2-vae.safetensors"
+    },
+    "class_type": "VAELoader",
+    "_meta": {
+      "title": "加载VAE"
+    }
+  },
+  "5": {
+    "inputs": {
+      "conditioning": [
+        "19",
+        0
+      ],
+      "latent": [
+        "13",
+        0
+      ]
+    },
+    "class_type": "ReferenceLatent",
+    "_meta": {
+      "title": "参考Latent"
+    }
+  },
+  "6": {
+    "inputs": {
+      "noise_seed": 217742615722421
+    },
+    "class_type": "RandomNoise",
+    "_meta": {
+      "title": "随机噪波"
+    }
+  },
+  "7": {
+    "inputs": {
+      "width": [
+        "14",
+        0
+      ],
+      "height": [
+        "14",
+        1
+      ],
+      "batch_size": 1
+    },
+    "class_type": "EmptySD3LatentImage",
+    "_meta": {
+      "title": "空Latent图像（SD3）"
+    }
+  },
+  "8": {
+    "inputs": {
+      "sampler_name": "euler"
+    },
+    "class_type": "KSamplerSelect",
+    "_meta": {
+      "title": "K采样器选择"
+    }
+  },
+  "9": {
+    "inputs": {
+      "noise": [
+        "6",
+        0
+      ],
+      "guider": [
+        "20",
+        0
+      ],
+      "sampler": [
+        "8",
+        0
+      ],
+      "sigmas": [
+        "1",
+        0
+      ],
+      "latent_image": [
+        "7",
+        0
+      ]
+    },
+    "class_type": "SamplerCustomAdvanced",
+    "_meta": {
+      "title": "自定义采样器（高级）"
+    }
+  },
+  "10": {
+    "inputs": {
+      "samples": [
+        "9",
+        0
+      ],
+      "vae": [
+        "3",
+        0
+      ]
+    },
+    "class_type": "VAEDecode",
+    "_meta": {
+      "title": "VAE解码"
+    }
+  },
+  "13": {
+    "inputs": {
+      "pixels": [
+        "44",
+        0
+      ],
+      "vae": [
+        "3",
+        0
+      ]
+    },
+    "class_type": "VAEEncode",
+    "_meta": {
+      "title": "VAE编码"
+    }
+  },
+  "14": {
+    "inputs": {
+      "image": [
+        "44",
+        0
+      ]
+    },
+    "class_type": "GetImageSize+",
+    "_meta": {
+      "title": "🔧 Get Image Size"
+    }
+  },
+  "16": {
+    "inputs": {
+      "unet_name": "flux2.0/flux-2-klein-9b-fp8.safetensors",
+      "weight_dtype": "fp8_e4m3fn"
+    },
+    "class_type": "UNETLoader",
+    "_meta": {
+      "title": "UNet加载器"
+    }
+  },
+  "17": {
+    "inputs": {
+      "filename_prefix": "ComfyUI",
+      "images": [
+        "62",
+        0
+      ]
+    },
+    "class_type": "SaveImage",
+    "_meta": {
+      "title": "保存图像"
+    }
+  },
+  "19": {
+    "inputs": {
+      "guidance": 1,
+      "conditioning": [
+        "22",
+        0
+      ]
+    },
+    "class_type": "FluxGuidance",
+    "_meta": {
+      "title": "Flux引导"
+    }
+  },
+  "20": {
+    "inputs": {
+      "model": [
+        "2",
+        0
+      ],
+      "conditioning": [
+        "5",
+        0
+      ]
+    },
+    "class_type": "BasicGuider",
+    "_meta": {
+      "title": "基本引导器"
+    }
+  },
+  "22": {
+    "inputs": {
+      "text": [
+        "60",
+        0
+      ],
+      "clip": [
+        "61",
+        0
+      ]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": {
+      "title": "CLIP文本编码"
+    }
+  },
+  "26": {
+    "inputs": {
+      "image": "clipspace/clipspace-painted-masked-1781655058499.png [input]"
+    },
+    "class_type": "LoadImage",
+    "_meta": {
+      "title": "加载图像"
+    }
+  },
+  "31": {
+    "inputs": {
+      "image": [
+        "26",
+        0
+      ]
+    },
+    "class_type": "easy imageSize",
+    "_meta": {
+      "title": "图像尺寸"
+    }
+  },
+  "32": {
+    "inputs": {
+      "aspect_ratio": "custom",
+      "proportional_width": [
+        "31",
+        0
+      ],
+      "proportional_height": [
+        "31",
+        1
+      ],
+      "fit": "letterbox",
+      "method": "lanczos",
+      "round_to_multiple": "8",
+      "scale_to_side": "None",
+      "scale_to_length": 1024,
+      "background_color": "#000000",
+      "image": [
+        "26",
+        0
+      ],
+      "mask": [
+        "37",
+        0
+      ]
+    },
+    "class_type": "LayerUtility: ImageScaleByAspectRatio V2",
+    "_meta": {
+      "title": "LayerUtility: ImageScaleByAspectRatio V2"
+    }
+  },
+  "33": {
+    "inputs": {
+      "masks": [
+        "26",
+        1
+      ]
+    },
+    "class_type": "Mask Fill Holes",
+    "_meta": {
+      "title": "Mask Fill Holes"
+    }
+  },
+  "36": {
+    "inputs": {
+      "masks": [
+        "33",
+        0
+      ]
+    },
+    "class_type": "Convert Masks to Images",
+    "_meta": {
+      "title": "Convert Masks to Images"
+    }
+  },
+  "37": {
+    "inputs": {
+      "method": "intensity",
+      "image": [
+        "39",
+        0
+      ]
+    },
+    "class_type": "Image To Mask",
+    "_meta": {
+      "title": "Image To Mask"
+    }
+  },
+  "39": {
+    "inputs": {
+      "upscale_method": "nearest-exact",
+      "width": [
+        "31",
+        0
+      ],
+      "height": [
+        "31",
+        1
+      ],
+      "crop": "disabled",
+      "image": [
+        "36",
+        0
+      ]
+    },
+    "class_type": "ImageScale",
+    "_meta": {
+      "title": "缩放图像"
+    }
+  },
+  "44": {
+    "inputs": {
+      "mask_opacity": 1,
+      "mask_color": "FFFF00",
+      "pass_through": true,
+      "image": [
+        "32",
+        0
+      ],
+      "mask": [
+        "32",
+        1
+      ]
+    },
+    "class_type": "ImageAndMaskPreview",
+    "_meta": {
+      "title": "ImageAndMaskPreview"
+    }
+  },
+  "45": {
+    "inputs": {
+      "images": [
+        "44",
+        0
+      ]
+    },
+    "class_type": "PreviewImage",
+    "_meta": {
+      "title": "预览图像"
+    }
+  },
+  "53": {
+    "inputs": {
+      "rgthree_comparer": {
+        "images": [
+          {
+            "name": "A",
+            "selected": true,
+            "url": "/api/view?filename=rgthree.compare._temp_zuixa_00119_.png&type=temp&subfolder=&rand=0.7926013627811991"
+          },
+          {
+            "name": "B",
+            "selected": true,
+            "url": "/api/view?filename=rgthree.compare._temp_zuixa_00120_.png&type=temp&subfolder=&rand=0.43692946883795813"
+          }
+        ]
+      },
+      "image_a": [
+        "62",
+        0
+      ],
+      "image_b": [
+        "26",
+        0
+      ]
+    },
+    "class_type": "Image Comparer (rgthree)",
+    "_meta": {
+      "title": "Image Comparer (rgthree)"
+    }
+  },
+  "60": {
+    "inputs": {
+      "text": "补充遮罩区补充遮罩区域内的头发，头发填满遮罩区域。发际线往下挡住额头"
+    },
+    "class_type": "JjkText",
+    "_meta": {
+      "title": "Text"
+    }
+  },
+  "61": {
+    "inputs": {
+      "clip_name": "qwen_3_8b_fp8mixed.safetensors",
+      "type": "flux2",
+      "device": "default"
+    },
+    "class_type": "CLIPLoader",
+    "_meta": {
+      "title": "加载CLIP"
+    }
+  },
+  "62": {
+    "inputs": {
+      "method": "mkl",
+      "strength": 1,
+      "multithread": true,
+      "image_ref": [
+        "26",
+        0
+      ],
+      "image_target": [
+        "10",
+        0
+      ]
+    },
+    "class_type": "ColorMatch",
+    "_meta": {
+      "title": "Color Match"
+    }
+  }
+}
@@ -1,6 +1,6 @@
 """旷视五接口 — worker 侧（高性能 GPU 后端）。

-接口 1（四庭七眼测量）已为**真实算法实现**（见 face_analysis 包）；接口 2~5 仍为 Mock。
+接口 1 等算法在 worker；接口4（用户特征）已迁到网关本机（调豆包），worker 同路径只返回明确错误、无 Mock。
 拆分架构：worker 跑算法、返回 `annotated_image_base64`（不落盘、不拼 URL，由网关完成）。
 worker 对 `/api/*` 校验内网鉴权头 `X-Internal-Token`，`/health` 供网关探测不校验。
 """
@@ -138,7 +138,8 @@ app = FastAPI(
 app.mount("/static", StaticFiles(directory="static"), name="static")

 # 不校验鉴权的路径前缀（供网关探测 / 文档 / 静态）
-_AUTH_EXEMPT = ("/health", "/docs", "/openapi.json", "/redoc", "/static", "/api/v1/debug")
+_AUTH_EXEMPT = ("/health", "/docs", "/openapi.json", "/redoc", "/static",
+               "/api/v1/debug", "/api/v1/redraw")


@app.middleware("http")
@@ -361,21 +362,36 @@ def _run_face_measure_data(image, variant="v1"):
        logger.warning("头发/耳朵分割失败，回退方案A：%s", seg_e)

    result = measure_face(landmarks, hair_mask, w, h, head_pose=head_pose)
+    discarded = result.hairline_discarded
    data = result.to_response()
+    vd = result.vertical
    if variant == "v6":
-        vd = result.vertical
-        base_px = vd["upper_court_px"] + vd["middle_court_px"] + vd["lower_court_px"]
-        # 接口6 是三庭：去掉顶庭相关字段（top_court_cm / ratios.top_court / landmarks.hair_top）
-        data["four_courts"]["ratios"] = {
-            "upper_court": round(vd["upper_court_px"] / base_px, 3),
-            "middle_court": round(vd["middle_court_px"] / base_px, 3),
-            "lower_court": round(vd["lower_court_px"] / base_px, 3),
-        }
-        data["four_courts"].pop("top_court_cm", None)
-        data["face_total_height_cm"] = round(
-            result.upper_cm + result.middle_cm + result.lower_cm, 2)
-        # 注：landmarks.hair_top 保留返回（供前端/下游定位头顶），但顶庭数值、
-        # 占比、标注图仍按三庭处理，显示效果不变。
+        if discarded:
+            # 发际线弃用：接口6 的上庭也依赖发际线，一并置 null；只保留中/下庭。
+            base_px = vd["middle_court_px"] + vd["lower_court_px"]
+            data["four_courts"]["upper_court_cm"] = None
+            data["four_courts"]["ratios"] = {
+                "upper_court": None,
+                "middle_court": round(vd["middle_court_px"] / base_px, 3),
+                "lower_court": round(vd["lower_court_px"] / base_px, 3),
+            }
+            data["four_courts"].pop("top_court_cm", None)
+            data["face_total_height_cm"] = round(
+                result.middle_cm + result.lower_cm, 2)
+            data["landmarks"]["hairline"] = None
+        else:
+            base_px = vd["upper_court_px"] + vd["middle_court_px"] + vd["lower_court_px"]
+            # 接口6 是三庭：去掉顶庭相关字段（top_court_cm / ratios.top_court / landmarks.hair_top）
+            data["four_courts"]["ratios"] = {
+                "upper_court": round(vd["upper_court_px"] / base_px, 3),
+                "middle_court": round(vd["middle_court_px"] / base_px, 3),
+                "lower_court": round(vd["lower_court_px"] / base_px, 3),
+            }
+            data["four_courts"].pop("top_court_cm", None)
+            data["face_total_height_cm"] = round(
+                result.upper_cm + result.middle_cm + result.lower_cm, 2)
+            # 注：landmarks.hair_top 保留返回（供前端/下游定位头顶），但顶庭数值、
+            # 占比、标注图仍按三庭处理，显示效果不变。

    # 七眼段宽度（cm）。eye1=左耳外段 eye2=左脸颊 eye3=左眼 eye4=两眼间距 eye5=右眼 eye6=右脸颊 eye7=右耳外段。
    # eye2~eye6（5段）只用内部分点，接口1/6 共用；eye1/eye7 需耳朵分割端线，仅接口1 有。
@@ -390,11 +406,14 @@ def _run_face_measure_data(image, variant="v1"):
            data["seven_eyes"][f"eye{i + 2}"] = (
                None if (a is None or b is None) else round((b - a) / pc, 2))
        if variant != "v6":
-            # 接口1 额外算 eye1/eye7（左/右耳外段），需耳朵分割端线
+            # 接口1 额外算 eye1/eye7（左/右耳外段），需耳朵分割端线。
+            # 竖向范围：发际线弃用时用眉心做上界（hair_top 不可靠），否则用头顶。
            from face_analysis.annotation import _ear_edges_from_mask
+            top_y = (vd["brow_center"][1] if discarded
+                     else vd["hair_top"][1])
            head_l, head_r = _ear_edges_from_mask(
                ear_mask, hair_mask,
-                result.vertical["hair_top"][1], result.vertical["chin_tip"][1],
+                top_y, vd["chin_tip"][1],
                lcx, rcx, (lcx + rcx) / 2)
            data["seven_eyes"]["eye1"] = (
                None if (head_l is None) else round((lcx - head_l) / pc, 2))
@@ -717,7 +736,7 @@ async def hair_grow(
    hair_style: Optional[str] = Form(default=None, description="发型序号逗号分隔（必填），如 1,2,3。female:1-5 male:1-4"),
    beauty_enabled: bool = Form(default=False, description="是否开启美颜（本期不生效）"),
    use_mask: bool = Form(default=True, description="是否启用 inpaint 遮罩（测试对比用）。false 时用干净原图生成（空遮罩，不烧模板线）"),
-    prompt: str = Form(default="补充遮罩区域的头发，加一点美颜", description="ComfyUI 提示词，会替换工作流节点60的文本"),
+    prompt: str = Form(default="填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜", description="ComfyUI 提示词，会替换工作流节点60的文本"),
 ):
    # 1. gender 必填校验（非法/缺失 → 1004）
    if gender not in ("male", "female"):
@@ -770,117 +789,15 @@ async def hair_grow(


 # ---------------------------------------------------------------------------
-# 接口 7：C 端生发 v2（add_hair2.json 工作流）
+# 接口 7：C 端生发 v2 —— 已弃用（add_hair2.json 用 Klein-9b 大模型，会把常驻的
+# Klein-4b/Flux 挤出显存，导致接口2/3/5 耗时抖动；且业务已不再调用）。
+# 保留路由返回明确错误，避免老客户端拿到裸 404。
 # ---------------------------------------------------------------------------

-_WORKFLOW2_PATH = os.path.join(os.path.dirname(__file__), "add_hair2.json")
-
-
-@app.post(
-    "/api/v1/hair/grow-v2",
-    summary="接口7 C端生发 v2（add_hair2 工作流）",
-    tags=["生发"],
-    description=f"""
-输入用户正面照 + **性别** + **发型序号**，使用 add_hair2.json 工作流生成指定发际线类型的预览图与生发图。
-功能与接口2 完全一致，仅 ComfyUI 工作流不同。
-
-{_image_fields_desc}
-
-图片同时支持 `multipart/form-data` 文件上传（字段名 `image_file`）。
-
---
-
- **gender**（必填）：`male` / `female`。决定返回的贴图集合（female 5 张 / male 4 张）。
-  非法或缺失返回 `1004`。
- **hair_style**（必填）：`int`，发型序号。`female`：1=ellipse, 2=flower, 3=heart, 4=straight, 5=wave；
-  `male`：1=ellipse, 2=inverse_arc, 3=m, 4=straight。越界返回 `1007`。
- **beauty_enabled**：本期保留但不生效。
-
-`hairline_type` 取值：`ellipse` / `flower` / `heart` / `straight` / `wave`（female），
-`ellipse` / `m` / `straight` / `inverse_arc`（male）。
-""",
-    responses={
-        200: {
-            "description": "成功",
-            "content": {
-                "application/json": {
-                    "example": {
-                        "code": 0,
-                        "message": "success",
-                        "request_id": "mock-request-id",
-                        "data": {
-                            "results": [
-                                {"image_base64": "iVBORw0KGgo...", "hairline_type": "ellipse", "order": 1},
-                            ]
-                        },
-                    }
-                }
-            },
-        },
-        400: {
-            "description": "参数错误 / 图片识别失败",
-            "content": {
-                "application/json": {
-                    "examples": {
-                        "图片参数错误": {"value": {"code": 1007, "message": "图片参数错误：必须且只能传 image_file / image_url / image_base64 其中一个", "request_id": "x", "data": None}},
-                        "非正面照": {"value": {"code": 1003, "message": "角度问题，请上传正面照", "request_id": "x", "data": None}},
-                    }
-                }
-            },
-        },
-    },
-)
-async def hair_grow_v2(
-    image_file: Optional[UploadFile] = File(default=None, description="上传图片文件（JPG/PNG）"),
-    image_url: Optional[str] = Form(default=None, description="图片 URL"),
-    image_base64: Optional[str] = Form(default=None, description="图片 base64（需带 data:image/...;base64, 前缀）"),
-    gender: Optional[str] = Form(default=None, description="性别 male/female（必填）"),
-    hair_style: Optional[str] = Form(default=None, description="发型序号逗号分隔（必填），如 1,2,3。female:1-5 male:1-4"),
-    beauty_enabled: bool = Form(default=False, description="是否开启美颜（本期不生效）"),
-    use_mask: bool = Form(default=True, description="是否启用 inpaint 遮罩（测试对比用）。false 时用干净原图生成（空遮罩，不烧模板线）"),
-    prompt: str = Form(default="补充遮罩区域的头发，加一点美颜", description="ComfyUI 提示词，会替换工作流节点60的文本"),
-):
-    # 1. gender 必填校验（非法/缺失 → 1004）
-    if gender not in ("male", "female"):
-        return err(1004, "gender 必填且只能为 male / female")
-
-    # 2. hair_style 必填校验（解析逗号分隔，越界 → 1007）
-    max_styles = {"female": 5, "male": 4}[gender]
-    hair_styles = _parse_hair_styles(hair_style, max_styles)
-    if hair_styles is None:
-        return err(1007, f"hair_style 必填且为 1..{max_styles} 的整数（逗号分隔），收到 {hair_style!r}")
-
-    # 3. 三选一取图
-    raw, e = await resolve_image_bytes(image_file, image_url, image_base64)
-    if e is not None:
-        return e
-
-    image = cv2.imdecode(np.frombuffer(raw, np.uint8), cv2.IMREAD_COLOR)
-    if image is None:
-        return err(1008, "图片格式不支持（仅 JPG / PNG）")
-
-    try:
-        from fastapi.concurrency import run_in_threadpool
-        from hairline.service import generate_grow_results
-
-        # 预览 + 生发(ComfyUI) 都是阻塞且较慢，放线程池避免卡住事件循环
-        items = await run_in_threadpool(generate_grow_results, image, gender, use_mask, prompt, hair_styles, _WORKFLOW2_PATH)
-        if items is None:
-            return err(1001, "无法识别人像")
-
-        results = []
-        for p in items:
-            results.append({
-                "image_base64": _jpg_b64(p["image_bgr"]),               # 预览图 JPG
-                "grown_image_base64": (_png_to_jpg_b64(p["grown_png"])  # 生发图 JPG
-                                       if p["grown_png"] else None),
-                "hairline_type": p["hairline_type"],
-                "order": p["order"],
-            })
-        return ok({"results": results})
-    except Exception as ex:  # noqa: BLE001
-        logger.exception("接口7 处理异常")
-        return err(1007, f"处理失败：{ex}")
+@app.post("/api/v1/hair/grow-v2", include_in_schema=False, deprecated=True)
+async def hair_grow_v2():
+    """接口7 已弃用：请改用 /api/v1/hair/grow（接口2）。"""
+    return err(1007, "接口7（/api/v1/hair/grow-v2）已弃用，请使用 /api/v1/hair/grow")


 # ---------------------------------------------------------------------------
@@ -930,7 +847,7 @@ async def hair_grow_b(
    marked_image_url: Optional[str] = Form(default=None, description="划线图片 URL"),
    marked_image_base64: Optional[str] = Form(default=None, description="划线图片 base64"),
    use_mask: bool = Form(default=True, description="是否画发际线（测试对比用）。false 时跳过划线检测、直接送划线图"),
-    prompt: str = Form(default="补充遮罩区域的头发，加一点美颜", description="ComfyUI 提示词，会替换工作流节点60的文本"),
+    prompt: str = Form(default="填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜", description="ComfyUI 提示词，会替换工作流节点60的文本"),
 ):
    # 划线图三选一取图（只需这一张）
    marked_raw, e = await resolve_image_bytes(marked_image_file, marked_image_url, marked_image_base64)
@@ -968,56 +885,28 @@ async def hair_grow_b(

@app.post(
    "/api/v1/face/features",
-    summary="接口4 用户特征分析",
+    summary="接口4 用户特征分析（仅网关）",
    tags=["人脸分析"],
-    description=f"""
-输入用户照片，返回 N 个用户面部特征字段。
+    description="""
+**本接口不在 worker 实现。** 请调用网关（本机默认 `http://127.0.0.1:8080`）的同路径；
+网关本机调火山方舟豆包视觉模型，不转发到 worker。

-{_image_fields_desc}
-
-图片同时支持 `multipart/form-data` 文件上传（字段名 `image_file`）。
-
---
-
-由**火山方舟 豆包视觉模型**分析，返回**固定 6 个英文字段**。
-
-**返回格式**：`data.features` 为一个 **JSON 字符串**（不是对象），需要在客户端 `JSON.parse()` 后使用。
-
-| 字段 | 说明 |
-|------|------|
-| face_shape | 脸形（如"鹅蛋脸"） |
-| eyebrow_shape | 眉形（如"平眉"） |
-| facial_age | 面部年龄（区间，如"18-25岁"） |
-| dynamic_static_type | 动静类型（"静态型"/"动态型"） |
-| gender | 性别（"男"/"女"） |
-| gene_style | 基因风格（如"自然型"） |
-
-> 无人脸返回 `1001`。
+直接打 worker（如 `:8187`）会返回错误，避免误用假数据。
 """,
    responses={
        200: {
-            "description": "成功",
+            "description": "worker 不提供本接口",
            "content": {
                "application/json": {
                    "example": {
-                        "code": 0,
-                        "message": "success",
+                        "code": 1007,
+                        "message": "接口4 仅在网关实现，请访问网关（本机默认 :8080），worker 不提供本接口",
                        "request_id": "mock-request-id",
-                        "data": {
-                            "features": '{"face_shape":"鹅蛋脸","eyebrow_shape":"平眉","facial_age":"18-25岁","dynamic_static_type":"静态型","gender":"女","gene_style":"少年型"}',
-                        },
+                        "data": None,
                    }
                }
            },
        },
-        400: {
-            "description": "参数错误 / 图片识别失败",
-            "content": {
-                "application/json": {
-                    "example": {"code": 1001, "message": "无法识别人像", "request_id": "x", "data": None}
-                }
-            },
-        },
    },
 )
 async def face_features(
@@ -1025,16 +914,12 @@ async def face_features(
    image_url: Optional[str] = Form(default=None, description="图片 URL"),
    image_base64: Optional[str] = Form(default=None, description="图片 base64（需带 data:image/...;base64, 前缀）"),
 ):
-    # ⚠️ 接口4 已迁到**网关本机**实现（直接调豆包视觉模型，见 gateway/app.py）。
-    #    网关不会把本接口转发到 worker，故此处仅留 Mock 占位、保持 worker 无外网依赖。
-    features = json.dumps(
-        {
-            "face_shape": "鹅蛋脸", "eyebrow_shape": "平眉", "facial_age": "18-25岁",
-            "dynamic_static_type": "静态型", "gender": "女", "gene_style": "少年型",
-        },
-        ensure_ascii=False,
+    # 接口4 只在网关实现（gateway/app.py → face_features.analyze_features）。
+    # 不再返回 Mock 成功数据，避免本机打 :8187 时被假结果误导。
+    return err(
+        1007,
+        "接口4 仅在网关实现，请访问网关（本机默认 :8080），worker 不提供本接口",
    )
-    return ok({"features": features})


 # ---------------------------------------------------------------------------
@@ -1061,6 +946,8 @@ async def face_features(
  `female`：1=ellipse,2=flower,3=heart,4=straight,5=wave；`male`：1=ellipse,2=inverse_arc,3=m,4=straight。
 - 可选 `use_mask` / `prompt`：同接口2 的生发控制参数。
  注：生发黑模板固定取 `hairline_texture_black/`（middle 档），即三档叠图分别用各自贴图、但生发目标固定 middle。
+- 可选 `generate_grow_image`（默认 `true`）：是否生成生发效果图（ComfyUI 生发，全流程最耗时）。
+  `false` 时跳过生发，各发型 `grown_image_*` 恒为 `null`，仅返回三档发际线叠图与中心点，大幅降低耗时。

 **返回说明**：

@@ -1138,7 +1025,8 @@ async def hairline_generate(
    gender: Optional[str] = Form(default=None, description="性别 male/female（必填）"),
    hair_style: Optional[str] = Form(default=None, description="发型序号逗号分隔（必填，如 1,2,3）。female:1-5 male:1-4"),
    use_mask: bool = Form(default=True, description="生发是否启用 inpaint 遮罩（同接口2，测试对比用）"),
-    prompt: str = Form(default="补充遮罩区域的头发，加一点美颜", description="ComfyUI 提示词（同接口2），会替换工作流节点60的文本"),
+    prompt: str = Form(default="填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜", description="ComfyUI 提示词（同接口2），会替换工作流节点60的文本"),
+    generate_grow_image: bool = Form(default=True, description="是否生成生发效果图（ComfyUI 生发，最耗时）。默认 true 出图；false 时跳过生发，各发型 grown_image 恒为 null，仅返回三档发际线叠图与中心点"),
 ):
    if gender not in ("male", "female"):
        return err(1004, "gender 必填且只能为 male / female")
@@ -1162,7 +1050,8 @@ async def hairline_generate(
        from hairline.service import generate_hairline_pngs

        res = await run_in_threadpool(
-            generate_hairline_pngs, image, gender, hair_styles, use_mask, prompt)
+            generate_hairline_pngs, image, gender, hair_styles, use_mask, prompt,
+            generate_grow_image=generate_grow_image)
        if res is None:
            return err(1001, "无法识别人像")

@@ -1530,7 +1419,7 @@ async def hairline_grow_v2(
    inpainting_fill: int = Form(default=1, description="change_hair服务端重绘填充：0=保留原图 | 1=噪声 | 2=纯色 | 3=潜变量。默认 1"),
    mask_blur: int = Form(default=11, description="change_hair服务端遮罩边缘模糊像素，默认 11"),
    mask_dilate_scale: float = Form(default=1.0, description="change_hair服务端遮罩膨胀缩放，默认 1.0"),
-    comfyui_prompt: Optional[str] = Form(default=None, description="Flux-2 重绘提示词，None 用默认「补充遮罩区域的头发，加一点美颜」"),
+    comfyui_prompt: Optional[str] = Form(default=None, description="Flux-2 重绘提示词，None 用默认「填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜」"),
    beauty_alpha: float = Form(default=0.6, description="redraw_band 版 band 外的全脸美颜融入强度（0=band外无美颜纯用final，1≈整帧版），默认 0.6"),
    band_lo_mult: float = Form(default=0.5, description="重绘带外推倍率下限（相对 hairline_push_cm，内轮廓=0×、原外推线=1.0×），默认 0.5"),
    band_hi_mult: float = Form(default=1.5, description="重绘带外推倍率上限（相对 hairline_push_cm），默认 1.5"),
@@ -1677,6 +1566,42 @@ async def hairline_grow_v2_final_v2(
    return await _run_v2_final(image_file, image_url, image_base64, hairline_id, "接口12finalv2")


+# ---------------------------------------------------------------------------
+# 重绘端点（替代 local_test /api/generate）
+# ---------------------------------------------------------------------------
+
+@app.post(
+    "/api/v1/redraw",
+    summary="ComfyUI 重绘",
+    tags=["重绘"],
+    description="""
+传入人物图片 + 遮罩图片，直接调 ComfyUI（0716add-hair 工作流）执行局部重绘。
+替代原 local_test :8899 的 /api/generate 接口。
+
+**遮罩图片格式**：支持红色遮罩(R=255)、白色遮罩(R=G=B=255)、Alpha遮罩(A=255)，服务取所有通道最大值。
+**遮罩区域**表示需要重绘的部分，非遮罩区域保持原图不变。
+""",
+)
+async def api_redraw(
+    image_file: UploadFile = File(..., description="人物图片（JPG/PNG）"),
+    mask_file: UploadFile = File(..., description="遮罩图片（PNG，支持红/白/alpha 格式）"),
+    prompt: str = Form(default="填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜",
+                       description="ComfyUI 提示词"),
+):
+    image_bytes = await image_file.read()
+    mask_bytes = await mask_file.read()
+    from fastapi.concurrency import run_in_threadpool
+    from hairline.redraw import run_redraw
+    try:
+        png_bytes = await run_in_threadpool(
+            run_redraw, image_bytes, mask_bytes, prompt)
+    except Exception as e:  # noqa: BLE001
+        logger.warning("重绘失败: %s", e)
+        return err(500, f"重绘失败: {e}")
+    b64 = base64.b64encode(png_bytes).decode()
+    return ok({"image_base64": f"data:image/png;base64,{b64}"})
+
+
 # ---------------------------------------------------------------------------
 # 调试：下载后端日志（接口11 遮罩计算全过程）
 # ---------------------------------------------------------------------------
@@ -1705,6 +1630,84 @@ async def download_hairline_log(rid: Optional[str] = None, tail: int = 500):
    return PlainTextResponse("".join(lines), media_type="text/plain; charset=utf-8")


+# ---------------------------------------------------------------------------
+# 调试：MediaPipe 脸型分类（face/face_shape_classifier.py，非接口4 豆包）
+# ---------------------------------------------------------------------------
+
+@app.post(
+    "/api/v1/debug/face-shape",
+    summary="调试 单张脸型分类（MediaPipe）",
+    tags=["调试"],
+    description="""
+离线脸型分类调试接口（`face/face_shape_classifier.py`），**不是**接口4 的豆包视觉分析。
+
+上传正面照 → MediaPipe 468 点 → 7 类脸型评分 + 特征标注图。
+""",
+    include_in_schema=True,
+)
+async def debug_face_shape(
+    image_file: Optional[UploadFile] = File(default=None, description="上传图片文件（JPG/PNG）"),
+    image_url: Optional[str] = Form(default=None, description="图片 URL"),
+    image_base64: Optional[str] = Form(default=None, description="图片 base64"),
+):
+    raw, e = await resolve_image_bytes(image_file, image_url, image_base64)
+    if e:
+        return e
+    try:
+        nparr = np.frombuffer(raw, np.uint8)
+        bgr = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+        if bgr is None:
+            return err(1008, "图片格式不支持（仅 JPG / PNG）")
+    except Exception:  # noqa: BLE001
+        return err(1008, "图片格式不支持（仅 JPG / PNG）")
+
+    def _jsonable(obj):
+        if isinstance(obj, dict):
+            return {k: _jsonable(v) for k, v in obj.items()}
+        if isinstance(obj, (list, tuple)):
+            return [_jsonable(v) for v in obj]
+        if hasattr(obj, "item"):
+            return obj.item()
+        if isinstance(obj, (float, int, str, bool)) or obj is None:
+            return obj
+        return obj
+
+    from fastapi.concurrency import run_in_threadpool
+
+    try:
+        from face.face_shape_classifier import classify_from_image
+
+        result = await run_in_threadpool(
+            classify_from_image, bgr, True, True,
+        )
+    except ValueError as ex:
+        return err(1001, str(ex) or "无法识别人像")
+    except Exception as ex:  # noqa: BLE001
+        return err(1007, f"脸型分类失败：{ex}")
+
+    details = result.get("details") or {}
+    ranked = [
+        {"shape": name, "score": round(float(score), 2)}
+        for name, score in (details.get("ranked") or [])
+    ]
+    annotated = result.pop("annotated", None)
+    h, w = bgr.shape[:2]
+    data = {
+        "face_shape": result["face_shape"],
+        "display": result["display"],
+        "confidence": round(float(result["confidence"]), 4),
+        "is_mixed": bool(details.get("is_mixed")),
+        "second_shape": details.get("second_shape"),
+        "score_gap": round(float(details["score_gap"]), 2) if details.get("score_gap") is not None else None,
+        "ranked": ranked,
+        "features": _jsonable(result.get("features") or {}),
+        "zscores": _jsonable(details.get("zscores") or {}),
+        "image_size": {"width": w, "height": h},
+        "annotated_image_base64": _jpg_b64(annotated) if annotated is not None else None,
+    }
+    return ok(data)
+
+
 # ---------------------------------------------------------------------------
 # 健康检查
 # ---------------------------------------------------------------------------
@@ -0,0 +1,157 @@
+#!/usr/bin/env python3
+"""9B vs 4B 模型对比：3 张图 × 5 种发型 = 15 张对比图。
+
+第一次运行：PLAN_TAG=A  （9B 模型，需先切工作流到 .bak）
+第二次运行：PLAN_TAG=Bplus（4B 模型，需切回 4B 工作流）
+"""
+import json, os, time, base64, subprocess, urllib.request
+from datetime import datetime
+
+API_BASE = "http://127.0.0.1:8187"
+TOKEN = "dev-shared-secret-2026"
+IMAGES = [
+    "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+]
+PLAN_TAG = os.getenv("PLAN_TAG", "A")
+OUT_DIR = f"/home/ubuntu/hair/benchmark_out/9b_vs_4b_{PLAN_TAG}"
+RESULT_JSON = os.path.join(OUT_DIR, "results.json")
+REPORT_HTML = os.path.join(OUT_DIR, "report.html")
+HAIR_STYLES = [(1, "椭圆"), (2, "花瓣"), (3, "心形"), (4, "直线"), (5, "波浪")]
+WARMUP = os.getenv("WARMUP", "1") == "1"  # 第一次调用做预热
+
+os.makedirs(OUT_DIR, exist_ok=True)
+
+
+def get_vram():
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=memory.used,memory.free",
+             "--format=csv,noheader,nounits"], text=True, timeout=5).strip()
+        used, free = out.split(",")
+        return int(used.strip()), int(free.strip())
+    except Exception:
+        return -1, -1
+
+
+def _multipart(fields, files=None):
+    boundary = "----CmpTest" + str(int(time.time() * 1000))
+    parts = []
+    for k, v in fields.items():
+        parts.append(f"--{boundary}\r\n".encode())
+        parts.append(f'Content-Disposition: form-data; name="{k}"\r\n\r\n'.encode())
+        parts.append(str(v).encode())
+        parts.append(b"\r\n")
+    if files:
+        for field_name, (filename, data, mime) in files.items():
+            parts.append(f"--{boundary}\r\n".encode())
+            parts.append(f'Content-Disposition: form-data; name="{field_name}"; filename="{filename}"\r\n'.encode())
+            parts.append(f"Content-Type: {mime}\r\n\r\n".encode())
+            parts.append(data)
+            parts.append(b"\r\n")
+    parts.append(f"--{boundary}--\r\n".encode())
+    return b"".join(parts), boundary
+
+
+def call_iface2(image_path, hair_style):
+    with open(image_path, "rb") as f:
+        img_data = f.read()
+    fields = {"gender": "female", "hair_style": str(hair_style),
+              "prompt": "填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜"}
+    body, boundary = _multipart(
+        fields, {"image_file": (os.path.basename(image_path), img_data, "image/jpeg")})
+    req = urllib.request.Request(f"{API_BASE}/api/v1/hair/grow", data=body, method="POST")
+    req.add_header("Content-Type", f"multipart/form-data; boundary={boundary}")
+    req.add_header("X-Internal-Token", TOKEN)
+    t0 = time.time()
+    try:
+        with urllib.request.urlopen(req, timeout=600) as resp:
+            data = json.loads(resp.read())
+            return data, time.time() - t0, None
+    except Exception as e:
+        return {}, time.time() - t0, str(e)
+
+
+def extract_image(data, img_idx, hs):
+    """从响应提取 grown_image_base64 并保存为 jpg。"""
+    if not data or not data.get("data"):
+        return None
+    items = data["data"].get("results") or []
+    if not items:
+        return None
+    b64 = items[0].get("grown_image_base64") or ""
+    if b64.startswith("data:"):
+        b64 = b64.split(",", 1)[1]
+    if not b64:
+        return None
+    fname = f"img{img_idx}_style{hs}.jpg"
+    path = os.path.join(OUT_DIR, fname)
+    with open(path, "wb") as f:
+        f.write(base64.b64decode(b64))
+    return path
+
+
+def run_test():
+    results = []
+    print(f"9B vs 4B 对比测试 — Plan={PLAN_TAG}")
+    print(f"图片: {[os.path.basename(p) for p in IMAGES]}")
+    print(f"发型: {[(s,n) for s,n in HAIR_STYLES]}")
+    print(f"总调用: {len(IMAGES)*len(HAIR_STYLES)}\n")
+
+    # 预热请求（避免第一次冷启动计入统计）
+    if WARMUP:
+        print("[warmup] 预热请求 (girl13, style=1) ...")
+        t0 = time.time()
+        _, warmup_time, _ = call_iface2(IMAGES[2], 1)
+        print(f"  warmup: {warmup_time:.1f}s\n")
+
+    for img_idx, img_path in enumerate(IMAGES, start=1):
+        img_name = os.path.basename(img_path)
+        for hs, hs_name in HAIR_STYLES:
+            print(f"[img{img_idx}/{len(IMAGES)}] {img_name} style={hs}({hs_name}) ...")
+
+            vram_before, _ = get_vram()
+            data, elapsed, error = call_iface2(img_path, hs)
+            vram_after, _ = get_vram()
+
+            code = data.get("code", -1) if data else -1
+            ok = (code == 0)
+            img_saved = extract_image(data, img_idx, hs) if ok else None
+
+            record = {
+                "plan": PLAN_TAG,
+                "image_idx": img_idx,
+                "image_name": img_name,
+                "image_path": img_path,
+                "hair_style": hs,
+                "hair_style_name": hs_name,
+                "timestamp": datetime.now().strftime("%H:%M:%S"),
+                "elapsed_s": round(elapsed, 2),
+                "success": ok,
+                "code": code,
+                "error": error,
+                "vram_before_mb": vram_before,
+                "vram_after_mb": vram_after,
+                "vram_delta_mb": vram_after - vram_before,
+                "saved_image_path": img_saved,
+            }
+            results.append(record)
+
+            status = "✓" if ok else "✗"
+            print(f"  → {status} code={code} time={elapsed:.1f}s "
+                  f"vram={vram_before}→{vram_after}MB (Δ{vram_after-vram_before:+d}) "
+                  f"img={'saved' if img_saved else 'none'}")
+
+            with open(RESULT_JSON, "w") as f:
+                json.dump(results, f, indent=2, ensure_ascii=False)
+
+    print(f"\n✅ 完成！{RESULT_JSON}")
+    print(f"📊 成功: {sum(1 for r in results if r['success'])}/{len(results)}")
+    if any(r['success'] for r in results):
+        avg = sum(r['elapsed_s'] for r in results if r['success']) / sum(1 for r in results if r['success'])
+        print(f"⏱  平均: {avg:.2f}s")
+
+
+if __name__ == "__main__":
+    run_test()
@@ -0,0 +1,450 @@
+#!/usr/bin/env python3
+"""基准测试：19张图片 × 5种发际线，记录每步耗时和显存变化。
+
+用法： python3 benchmark_grow.py
+输出： benchmark_results.json + benchmark_report.html
+"""
+import json, os, time, base64, subprocess, re, glob, shutil
+from datetime import datetime
+from pathlib import Path
+import urllib.request, urllib.error
+
+API_URL = "http://127.0.0.1:8187/api/v1/hair/grow"
+TOKEN = "dev-shared-secret-2026"
+IMG_DIR = "/home/ubuntu/hair/image/girl_img"
+OUT_DIR = "/home/ubuntu/hair/benchmark_out"
+RESULT_JSON = os.path.join(OUT_DIR, "benchmark_results.json")
+REPORT_HTML = os.path.join(OUT_DIR, "benchmark_report.html")
+
+HAIRLINES = [
+    ("1", "ellipse",  "椭圆形"),
+    ("2", "flower",   "花瓣形"),
+    ("3", "heart",    "心形"),
+    ("4", "straight", "直线形"),
+    ("5", "wave",     "波浪形"),
+]
+
+os.makedirs(OUT_DIR, exist_ok=True)
+
+
+# ── 工具函数 ──────────────────────────────────────────────
+
+def get_vram():
+    """返回 (used_MB, free_MB)"""
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=memory.used,memory.free",
+             "--format=csv,noheader,nounits"], text=True, timeout=5
+        ).strip()
+        used, free = out.split(",")
+        return int(used.strip()), int(free.strip())
+    except Exception:
+        return -1, -1
+
+
+def get_gpu_procs():
+    """返回各进程显存占用 dict"""
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-compute-apps=pid,used_memory",
+             "--format=csv,noheader,nounits"], text=True, timeout=5
+        ).strip()
+        procs = {}
+        for line in out.splitlines():
+            parts = line.split(",")
+            if len(parts) >= 2:
+                procs[parts[0].strip()] = int(parts[1].strip())
+        return procs
+    except Exception:
+        return {}
+
+
+def read_worker_log_tail(n=50):
+    """读取 worker.log 最后 n 行"""
+    log_path = "/home/ubuntu/hair/worker.log"
+    try:
+        result = subprocess.run(["tail", "-n", str(n), log_path],
+                                 capture_output=True, text=True, timeout=5)
+        return result.stdout
+    except Exception:
+        return ""
+
+
+def call_api(image_path, hair_style_value):
+    """调用接口2，返回 (json_dict, elapsed_sec, error_or_None)"""
+    import mimetypes
+    boundary = "----BenchmarkBoundary" + str(int(time.time()*1000))
+    filename = os.path.basename(image_path)
+    mime = mimetypes.guess_type(image_path)[0] or "image/jpeg"
+
+    with open(image_path, "rb") as f:
+        img_data = f.read()
+
+    body_parts = []
+    body_parts.append(f"--{boundary}\r\n".encode())
+    body_parts.append(f'Content-Disposition: form-data; name="image_file"; filename="{filename}"\r\n'.encode())
+    body_parts.append(f"Content-Type: {mime}\r\n\r\n".encode())
+    body_parts.append(img_data)
+    body_parts.append(f"\r\n--{boundary}\r\n".encode())
+    body_parts.append(b'Content-Disposition: form-data; name="gender"\r\n\r\n')
+    body_parts.append(b"female")
+    body_parts.append(f"\r\n--{boundary}\r\n".encode())
+    body_parts.append(b'Content-Disposition: form-data; name="hair_style"\r\n\r\n')
+    body_parts.append(hair_style_value.encode())
+    body_parts.append(f"\r\n--{boundary}\r\n".encode())
+    body_parts.append(b'Content-Disposition: form-data; name="use_mask"\r\n\r\n')
+    body_parts.append(b"0")
+    body_parts.append(f"\r\n--{boundary}\r\n".encode())
+    body_parts.append(b'Content-Disposition: form-data; name="prompt"\r\n\r\n')
+    body_parts.append(b"\xe5\xa1\xab\xe5\x85\x85\xe9\x81\xae\xe7\xbd\xa9\xe5\x8c\xba\xe5\x9f\x9f\xe7\x9a\x84\xe5\xa4\xb4\xe5\x8f\x91\xef\xbc\x8c\xe7\x9a\xae\xe8\x82\xa4\xe5\x8a\xa0\xe4\xb8\x80\xe7\x82\xb9\xe7\xa3\xa8\xe7\x9a\xae")
+    body_parts.append(f"\r\n--{boundary}--\r\n".encode())
+    body = b"".join(body_parts)
+
+    req = urllib.request.Request(API_URL, data=body, method="POST")
+    req.add_header("Content-Type", f"multipart/form-data; boundary={boundary}")
+    req.add_header("X-Internal-Token", TOKEN)
+
+    t0 = time.time()
+    try:
+        with urllib.request.urlopen(req, timeout=600) as resp:
+            raw = resp.read()
+            elapsed = time.time() - t0
+            data = json.loads(raw)
+            return data, elapsed, None
+    except urllib.error.HTTPError as e:
+        elapsed = time.time() - t0
+        try:
+            data = json.loads(e.read())
+        except Exception:
+            data = {"error": str(e)}
+        return data, elapsed, f"HTTP {e.code}"
+    except Exception as e:
+        elapsed = time.time() - t0
+        return {}, elapsed, str(e)
+
+
+def parse_step_timing(log_text):
+    """从 worker.log 文本中解析步骤耗时"""
+    timing = {}
+    for line in log_text.splitlines():
+        # 匹配 "步骤1 遮罩完成 耗时=949ms"
+        m = re.search(r"步骤(\d+)\s+\S+\s+耗时=(\d+)ms", line)
+        if m:
+            timing[f"step{m.group(1)}_ms"] = int(m.group(2))
+        # 匹配 "人脸检出 px_per_cm=30.669 图尺寸=813x967"
+        if "人脸检出" in line:
+            m2 = re.search(r"px_per_cm=([\d.]+)", line)
+            if m2:
+                timing["px_per_cm"] = float(m2.group(1))
+        # 匹配 "头发分割完成"
+        if "头发分割完成" in line:
+            timing["hair_seg"] = True
+        # 匹配 "换发型图失败"
+        if "换发型图失败" in line or "换发型服务不可达" in line:
+            timing["swap_error"] = line.strip()[-100:]
+        # 匹配 "重绘" 相关
+        if "重绘" in line and ("完成" in line or "失败" in line):
+            timing["redraw_status"] = "完成" if "完成" in line else "失败"
+    return timing
+
+
+# ── 主流程 ────────────────────────────────────────────────
+
+def run_benchmark():
+    images = sorted(glob.glob(os.path.join(IMG_DIR, "girl*.jpg")))
+    print(f"找到 {len(images)} 张图片")
+
+    # 加载已有结果（支持断点续跑）
+    results = []
+    if os.path.exists(RESULT_JSON):
+        with open(RESULT_JSON) as f:
+            results = json.load(f)
+        print(f"已有 {len(results)} 条记录，继续未完成的测试")
+
+    total_calls = len(images) * len(HAIRLINES)
+    done = len(results)
+    print(f"总计 {total_calls} 次调用，已完成 {done}，剩余 {total_calls - done}")
+
+    log_offset = 0
+    try:
+        log_offset = subprocess.check_output(["wc", "-l", "/home/ubuntu/hair/worker.log"],
+                                              text=True, timeout=5).split()[0]
+        log_offset = int(log_offset)
+    except Exception:
+        pass
+
+    for img_idx, img_path in enumerate(images):
+        img_name = os.path.basename(img_path)
+
+        # 跳过已完成的图片
+        img_results = [r for r in results if r["image"] == img_name]
+        if len(img_results) >= len(HAIRLINES):
+            print(f"[{img_idx+1}/{len(images)}] {img_name} 已完成，跳过")
+            continue
+
+        for hl_id, hl_key, hl_label in HAIRLINES:
+            # 跳过已完成的
+            existing = [r for r in results if r["image"] == img_name and r["hairline_id"] == hl_id]
+            if existing:
+                continue
+
+            print(f"\n[{img_idx+1}/{len(images)}] {img_name} → {hl_label}({hl_id}) ...")
+
+            # 记录 worker.log 行数
+            try:
+                log_before = int(subprocess.check_output(
+                    ["wc", "-l", "/home/ubuntu/hair/worker.log"], text=True, timeout=5).split()[0])
+            except Exception:
+                log_before = 0
+
+            # VRAM before
+            vram_before, vram_free_before = get_vram()
+            procs_before = get_gpu_procs()
+            t_start = time.time()
+
+            # 调用API
+            api_result, elapsed, error = call_api(img_path, hl_id)
+
+            t_end = time.time()
+
+            # VRAM after
+            vram_after, vram_free_after = get_vram()
+            procs_after = get_gpu_procs()
+
+            # 读取新日志
+            try:
+                log_diff = subprocess.check_output(
+                    ["tail", "-n", "+{}".format(log_before + 1), "/home/ubuntu/hair/worker.log"],
+                    text=True, timeout=5)
+            except Exception:
+                log_diff = ""
+
+            step_timing = parse_step_timing(log_diff)
+
+            # 提取结果图片
+            preview_b64 = ""
+            grown_b64 = ""
+            api_code = api_result.get("code", -1)
+            api_msg = api_result.get("message", "")
+            api_results = api_result.get("data", {}).get("results", [])
+            if api_results:
+                r0 = api_results[0]
+                preview_b64 = r0.get("image_base64", "")
+                grown_b64 = r0.get("grown_image_base64", "")
+
+            # 保存缩略图
+            thumb_dir = os.path.join(OUT_DIR, "thumbs")
+            os.makedirs(thumb_dir, exist_ok=True)
+            if grown_b64:
+                grown_bytes = base64.b64decode(grown_b64)
+                thumb_path = os.path.join(thumb_dir, f"{img_name}_hl{hl_id}_grown.jpg")
+                with open(thumb_path, "wb") as f:
+                    f.write(grown_bytes)
+            if preview_b64:
+                preview_bytes = base64.b64decode(preview_b64)
+                thumb_path = os.path.join(thumb_dir, f"{img_name}_hl{hl_id}_preview.png")
+                with open(thumb_path, "wb") as f:
+                    f.write(preview_bytes)
+
+            record = {
+                "image": img_name,
+                "image_idx": img_idx + 1,
+                "hairline_id": hl_id,
+                "hairline_key": hl_key,
+                "hairline_label": hl_label,
+                "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
+                "total_time_s": round(elapsed, 2),
+                "api_code": api_code,
+                "api_msg": api_msg,
+                "error": error,
+                "has_preview": bool(preview_b64),
+                "has_grown": bool(grown_b64),
+                "preview_size": len(preview_b64),
+                "grown_size": len(grown_b64),
+                "vram_before_mb": vram_before,
+                "vram_after_mb": vram_after,
+                "vram_free_before_mb": vram_free_before,
+                "vram_free_after_mb": vram_free_after,
+                "vram_delta_mb": vram_after - vram_before,
+                "procs_before": procs_before,
+                "procs_after": procs_after,
+                "step_timing": step_timing,
+            }
+            results.append(record)
+            print(f"  → HTTP code={api_code} time={elapsed:.1f}s vram={vram_before}→{vram_after}MB "
+                  f"preview={'✓' if preview_b64 else '✗'} grown={'✓' if grown_b64 else '✗'}")
+
+            # 保存中间结果
+            with open(RESULT_JSON, "w") as f:
+                json.dump(results, f, indent=2, ensure_ascii=False)
+
+            # 生成中间报告
+            generate_html(results)
+
+    # 最终报告
+    generate_html(results)
+    print(f"\n✅ 完成！结果: {RESULT_JSON}")
+    print(f"📄 报告: {REPORT_HTML}")
+    print(f"📊 总调用: {len(results)}/{total_calls}")
+
+
+def generate_html(results):
+    """生成HTML报告"""
+    # 统计
+    total = len(results)
+    success = sum(1 for r in results if r["has_grown"])
+    failed = total - success
+    times = [r["total_time_s"] for r in results if r["has_grown"]]
+    avg_time = sum(times) / len(times) if times else 0
+    max_time = max(times) if times else 0
+    min_time = min(times) if times else 0
+
+    # 按发际线类型分组统计
+    hl_stats = {}
+    for r in results:
+        if r["has_grown"]:
+            hl = r["hairline_label"]
+            if hl not in hl_stats:
+                hl_stats[hl] = {"count": 0, "times": []}
+            hl_stats[hl]["count"] += 1
+            hl_stats[hl]["times"].append(r["total_time_s"])
+
+    # 按图片分组
+    img_groups = {}
+    for r in results:
+        img = r["image"]
+        if img not in img_groups:
+            img_groups[img] = []
+        img_groups[img].append(r)
+
+    # 生成VRAM变化数据
+    vram_data = [(i, r["vram_after_mb"]) for i, r in enumerate(results)]
+
+    html = f"""<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+<meta charset="UTF-8">
+<title>接口2 基准测试报告</title>
+<style>
+* {{ margin:0; padding:0; box-sizing:border-box; }}
+body {{ font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif; background:#1a1a2e; color:#e0e0e0; padding:20px; }}
+h1 {{ text-align:center; margin-bottom:20px; color:#00d4ff; }}
+.subtitle {{ text-align:center; color:#888; margin-bottom:30px; font-size:14px; }}
+.summary {{ display:grid; grid-template-columns:repeat(auto-fit,minmax(180px,1fr)); gap:15px; margin-bottom:30px; }}
+.card {{ background:#16213e; border-radius:12px; padding:20px; text-align:center; border:1px solid #333; }}
+.card .num {{ font-size:32px; font-weight:700; }}
+.card .label {{ font-size:12px; color:#888; margin-top:5px; }}
+.card.ok .num {{ color:#0f0; }}
+.card.err .num {{ color:#f44; }}
+.card.time .num {{ color:#00d4ff; }}
+table {{ width:100%; border-collapse:collapse; margin-bottom:30px; background:#16213e; border-radius:12px; overflow:hidden; }}
+th {{ background:#0f3460; padding:12px 8px; text-align:center; font-size:13px; color:#fff; }}
+td {{ padding:8px; text-align:center; border-bottom:1px solid #222; font-size:13px; }}
+tr:hover {{ background:#1a1a3e; }}
+.img-cell {{ text-align:left; }}
+.time-bar {{ display:inline-block; height:20px; background:linear-gradient(90deg,#0f3460,#00d4ff); border-radius:4px; vertical-align:middle; min-width:2px; }}
+.hl-badge {{ display:inline-block; padding:2px 8px; border-radius:10px; font-size:11px; font-weight:600; }}
+.hl-ellipse {{ background:#1b4332; color:#52b788; }}
+.hl-flower {{ background:#3a0ca3; color:#c77dff; }}
+.hl-heart {{ background:#6a040f; color:#ff6b6b; }}
+.hl-straight {{ background:#0077b6; color:#90e0ef; }}
+.hl-wave {{ background:#9d4edd; color:#e0aaff; }}
+.ok {{ color:#0f0; }} .fail {{ color:#f44; }}
+.vram-chart {{ margin:20px 0; }}
+.vram-bars {{ display:flex; align-items:flex-end; height:120px; gap:2px; padding:10px; background:#0d1117; border-radius:8px; }}
+.vram-bar {{ flex:1; background:linear-gradient(180deg,#00d4ff,#0f3460); border-radius:2px 2px 0 0; min-height:2px; position:relative; }}
+.vram-bar:hover::after {{ content:attr(data-val) 'MB'; position:absolute; bottom:100%; left:50%; transform:translateX(-50%); background:#333; padding:2px 6px; border-radius:4px; font-size:10px; white-space:nowrap; }}
+.section-title {{ font-size:18px; font-weight:600; margin:30px 0 15px; color:#00d4ff; border-bottom:1px solid #333; padding-bottom:10px; }}
+.hl-stats {{ display:grid; grid-template-columns:repeat(5,1fr); gap:15px; margin-bottom:20px; }}
+.hl-card {{ background:#16213e; border-radius:12px; padding:15px; text-align:center; }}
+.hl-card .avg {{ font-size:24px; font-weight:700; color:#00d4ff; }}
+.hl-card .minmax {{ font-size:11px; color:#888; margin-top:5px; }}
+.thumb {{ max-width:100px; max-height:100px; border-radius:4px; cursor:pointer; }}
+.thumb:hover {{ transform:scale(2); transition:transform 0.3s; }}
+</style>
+</head>
+<body>
+<h1>📊 接口2 基准测试报告</h1>
+<div class="subtitle">生成时间：{datetime.now().strftime('%Y-%m-%d %H:%M:%S')} ｜ {total} 次调用</div>
+
+<div class="summary">
+<div class="card ok"><div class="num">{success}</div><div class="label">成功</div></div>
+<div class="card err"><div class="num">{failed}</div><div class="label">失败</div></div>
+<div class="card time"><div class="num">{avg_time:.1f}s</div><div class="label">平均耗时</div></div>
+<div class="card time"><div class="num">{min_time:.1f}s</div><div class="label">最快</div></div>
+<div class="card time"><div class="num">{max_time:.1f}s</div><div class="label">最慢</div></div>
+</div>
+"""
+
+    # 按发际线类型统计
+    if hl_stats:
+        html += '<div class="section-title">按发际线类型统计</div><div class="hl-stats">'
+        for hl_label in ["椭圆形","花瓣形","心形","直线形","波浪形"]:
+            if hl_label in hl_stats:
+                s = hl_stats[hl_label]
+                times = s["times"]
+                avg = sum(times) / len(times)
+                mn, mx = min(times), max(times)
+                cls = hl_label
+                html += f'<div class="hl-card"><div class="hl-badge hl-{{cls}}">{hl_label}</div><div class="avg">{avg:.1f}s</div><div class="minmax">{mn:.1f}~{mx:.1f}s ({s["count"]}次)</div></div>'
+            else:
+                html += f'<div class="hl-card"><div class="hl-badge">{hl_label}</div><div class="avg">-</div><div class="minmax">未完成</div></div>'
+        html += '</div>'
+
+    # VRAM 变化图
+    if vram_data:
+        max_vram = max(v for _, v in vram_data if v > 0) or 1
+        html += '<div class="section-title">显存变化</div><div class="vram-chart"><div class="vram-bars">'
+        for i, (_, vram) in enumerate(vram_data):
+            if vram > 0:
+                h = int(vram / max_vram * 100)
+                html += f'<div class="vram-bar" style="height:{h}%" data-val="{vram}" title="第{i+1}次"></div>'
+            else:
+                html += f'<div class="vram-bar" style="height:0%" data-val="0"></div>'
+        html += '</div></div>'
+
+    # 详细结果表格
+    html += '<div class="section-title">详细结果</div><table><thead><tr>'
+    html += '<th>#</th><th>图片</th><th>发际线</th><th>总耗时</th><th>遮罩步骤</th>'
+    html += '<th>显存前</th><th>显存后</th><th>显存变化</th>'
+    html += '<th>预览图</th><th>生发图</th><th>状态</th>'
+    html += '</tr></thead><tbody>'
+
+    for i, r in enumerate(results):
+        hl_cls = r["hairline_key"]
+        status = '<span class="ok">✓ 成功</span>' if r["has_grown"] else f'<span class="fail">✗ {r.get("error","")}</span>'
+        step_ms = r.get("step_timing", {}).get("step1_ms", "")
+        step_str = f"{step_ms}ms" if step_ms else "-"
+        vram_d = r["vram_delta_mb"]
+        vram_d_str = f'<span style="color:{"#f44" if vram_d>0 else "#0f0"}">{"+" if vram_d>=0 else ""}{vram_d}</span>'
+
+        # 缩略图
+        thumb_grown = ""
+        if r["has_grown"]:
+            thumb_path = f"thumbs/{r['image']}_hl{r['hairline_id']}_grown.jpg"
+            if os.path.exists(os.path.join(OUT_DIR, thumb_path)):
+                thumb_grown = f'<img class="thumb" src="{thumb_path}">'
+
+        html += f"""<tr>
+<td>{i+1}</td>
+<td class="img-cell">{r['image']}</td>
+<td><span class="hl-badge hl-{hl_cls}">{r['hairline_label']}</span></td>
+<td><div class="time-bar" style="width:{min(r['total_time_s'],300)}px"></div> {r['total_time_s']:.1f}s</td>
+<td>{step_str}</td>
+<td>{r['vram_before_mb']}MB</td>
+<td>{r['vram_after_mb']}MB</td>
+<td>{vram_d_str}</td>
+<td>{'✓' if r['has_preview'] else '✗'}</td>
+<td>{thumb_grown if thumb_grown else ('✓' if r['has_grown'] else '✗')}</td>
+<td>{status}</td>
+</tr>"""
+
+    html += '</tbody></table>'
+    html += '</body></html>'
+
+    with open(REPORT_HTML, "w", encoding="utf-8") as f:
+        f.write(html)
+
+
+if __name__ == "__main__":
+    run_benchmark()
@@ -0,0 +1,201 @@
+#!/usr/bin/env python3
+"""接口2女性5种发型对比测试：Plan B vs Plan B+。
+
+测试 hair_style=1..5（椭圆/花瓣/心形/直线/波浪），每种发型一次，记录耗时+显存+生成图。
+"""
+import json, os, time, base64, subprocess, urllib.request, urllib.error
+from datetime import datetime
+
+API_BASE = "http://127.0.0.1:8187"
+TOKEN = "dev-shared-secret-2026"
+GIRL_IMG = "/home/ubuntu/hair/image/girl_img/girl13.jpg"
+PLAN_TAG = os.getenv("PLAN_TAG", "Bplus")  # Bplus / B
+OUT_DIR = f"/home/ubuntu/hair/benchmark_out/iface2_female_{PLAN_TAG}"
+RESULT_JSON = os.path.join(OUT_DIR, "results.json")
+REPORT_HTML = os.path.join(OUT_DIR, "report.html")
+
+HAIR_STYLE_NAMES = {
+    "1": "椭圆", "2": "花瓣", "3": "心形", "4": "直线", "5": "波浪",
+}
+
+os.makedirs(OUT_DIR, exist_ok=True)
+
+
+def get_vram():
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=memory.used,memory.free",
+             "--format=csv,noheader,nounits"], text=True, timeout=5).strip()
+        used, free = out.split(",")
+        return int(used.strip()), int(free.strip())
+    except Exception:
+        return -1, -1
+
+
+def _multipart(fields, files=None):
+    boundary = "----If2Test" + str(int(time.time() * 1000))
+    parts = []
+    for k, v in fields.items():
+        parts.append(f"--{boundary}\r\n".encode())
+        parts.append(f'Content-Disposition: form-data; name="{k}"\r\n\r\n'.encode())
+        parts.append(str(v).encode())
+        parts.append(b"\r\n")
+    if files:
+        for field_name, (filename, data, mime) in files.items():
+            parts.append(f"--{boundary}\r\n".encode())
+            parts.append(f'Content-Disposition: form-data; name="{field_name}"; filename="{filename}"\r\n'.encode())
+            parts.append(f"Content-Type: {mime}\r\n\r\n".encode())
+            parts.append(data)
+            parts.append(b"\r\n")
+    parts.append(f"--{boundary}--\r\n".encode())
+    return b"".join(parts), boundary
+
+
+def call_iface2_female(image_path, hair_style):
+    """接口2 女性 + 指定发型（走 swapHair 路径）"""
+    with open(image_path, "rb") as f:
+        img_data = f.read()
+    fields = {"gender": "female", "hair_style": str(hair_style),
+              "prompt": "填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜"}
+    body, boundary = _multipart(
+        fields, {"image_file": (os.path.basename(image_path), img_data, "image/jpeg")})
+    req = urllib.request.Request(f"{API_BASE}/api/v1/hair/grow", data=body, method="POST")
+    req.add_header("Content-Type", f"multipart/form-data; boundary={boundary}")
+    req.add_header("X-Internal-Token", TOKEN)
+    t0 = time.time()
+    try:
+        with urllib.request.urlopen(req, timeout=600) as resp:
+            data = json.loads(resp.read())
+            return data, time.time() - t0, None
+    except Exception as e:
+        return {}, time.time() - t0, str(e)
+
+
+def run_test():
+    results = []
+    print(f"接口2女性5种发型测试 — Plan={PLAN_TAG}")
+    print(f"图片：{GIRL_IMG}\n")
+
+    for hs in range(1, 6):
+        name = HAIR_STYLE_NAMES[str(hs)]
+        print(f"\n[{hs}/5] hair_style={hs} ({name}) ...")
+
+        vram_before, vram_free_before = get_vram()
+        data, elapsed, error = call_iface2_female(GIRL_IMG, hs)
+        vram_after, vram_free_after = get_vram()
+
+        code = data.get("code", -1) if data else -1
+        ok = (code == 0)
+
+        # 保存生成图（如果有）
+        img_path = None
+        if ok and data.get("data"):
+            try:
+                items = data["data"].get("results") or []
+                if items and isinstance(items, list):
+                    first = items[0]
+                    b64 = first.get("grown_image_base64") or ""
+                    if b64.startswith("data:"):
+                        b64 = b64.split(",", 1)[1]
+                    if b64:
+                        img_path = os.path.join(OUT_DIR, f"hair_style_{hs}_{name}.jpg")
+                        with open(img_path, "wb") as f:
+                            f.write(base64.b64decode(b64))
+            except Exception as e:
+                print(f"  保存图片失败: {e}")
+
+        record = {
+            "plan": PLAN_TAG,
+            "hair_style": hs,
+            "hair_style_name": name,
+            "timestamp": datetime.now().strftime("%H:%M:%S"),
+            "elapsed_s": round(elapsed, 2),
+            "success": ok,
+            "code": code,
+            "error": error,
+            "vram_before_mb": vram_before,
+            "vram_after_mb": vram_after,
+            "vram_delta_mb": vram_after - vram_before,
+            "image_path": img_path,
+        }
+        results.append(record)
+
+        status = "✓" if ok else "✗"
+        print(f"  → {status} code={code} time={elapsed:.1f}s "
+              f"vram={vram_before}→{vram_after}MB (Δ{vram_after-vram_before:+d}) "
+              f"img={'saved' if img_path else 'none'}")
+
+        with open(RESULT_JSON, "w") as f:
+            json.dump(results, f, indent=2, ensure_ascii=False)
+        generate_html(results)
+
+    generate_html(results)
+    print(f"\n✅ 完成！{RESULT_JSON}")
+    print(f"📄 报告：{REPORT_HTML}")
+
+
+def generate_html(results):
+    total = len(results)
+    success = sum(1 for r in results if r["success"])
+    times = [r["elapsed_s"] for r in results if r["success"]]
+    avg_time = sum(times) / len(times) if times else 0
+
+    html = f"""<!DOCTYPE html>
+<html lang="zh-CN"><head><meta charset="UTF-8"><title>接口2女5种发型 - Plan {PLAN_TAG}</title>
+<style>
+*{{margin:0;padding:0;box-sizing:border-box}}
+body{{font-family:sans-serif;background:#1a1a2e;color:#e0e0e0;padding:20px}}
+h1{{text-align:center;color:#00d4ff;margin-bottom:10px}}
+.subtitle{{text-align:center;color:#888;margin-bottom:30px;font-size:14px}}
+.summary{{display:grid;grid-template-columns:repeat(auto-fit,minmax(160px,1fr));gap:15px;margin-bottom:30px}}
+.card{{background:#16213e;border-radius:12px;padding:20px;text-align:center;border:1px solid #333}}
+.card .num{{font-size:28px;font-weight:700}}
+.card .label{{font-size:12px;color:#888;margin-top:5px}}
+.card.ok .num{{color:#0f0}} .card.err .num{{color:#f44}} .card.time .num{{color:#00d4ff}}
+table{{width:100%;border-collapse:collapse;margin-bottom:30px;background:#16213e;border-radius:12px;overflow:hidden}}
+th{{background:#0f3460;padding:10px 8px;text-align:center;font-size:13px;color:#fff}}
+td{{padding:8px;text-align:center;border-bottom:1px solid #222;font-size:13px}}
+tr:hover{{background:#1a1a3e}}
+.section-title{{font-size:18px;font-weight:600;margin:30px 0 15px;color:#00d4ff;border-bottom:1px solid #333;padding-bottom:10px}}
+.ok{{color:#0f0}} .fail{{color:#f44}}
+.time-bar{{display:inline-block;height:18px;background:linear-gradient(90deg,#0f3460,#00d4ff);border-radius:3px;vertical-align:middle;min-width:2px}}
+.gallery{{display:grid;grid-template-columns:repeat(auto-fit,minmax(200px,1fr));gap:15px}}
+.gallery img{{width:100%;border-radius:8px;border:1px solid #333}}
+.gallery .item{{text-align:center}}
+.gallery .cap{{margin-top:5px;font-size:12px;color:#888}}
+</style></head><body>
+<h1>📊 接口2女性5种发型测试报告</h1>
+<div class="subtitle">Plan {PLAN_TAG} ｜ {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} ｜ {total} 次调用</div>
+<div class="summary">
+<div class="card ok"><div class="num">{success}</div><div class="label">成功</div></div>
+<div class="card err"><div class="num">{total-success}</div><div class="label">失败</div></div>
+<div class="card time"><div class="num">{avg_time:.1f}s</div><div class="label">平均耗时</div></div>
+</div>
+<div class="section-title">详细结果</div>
+<table><thead><tr><th>发型</th><th>名称</th><th>耗时</th><th>显存前</th><th>显存后</th><th>Δ</th><th>状态</th></tr></thead><tbody>
+"""
+    for r in results:
+        status = '<span class="ok">✓</span>' if r["success"] else f'<span class="fail">✗ {str(r.get("error",""))[:30]}</span>'
+        html += f'<tr><td>style {r["hair_style"]}</td><td>{r["hair_style_name"]}</td>' \
+                f'<td><div class="time-bar" style="width:{min(r["elapsed_s"]*15,200)}px"></div> {r["elapsed_s"]:.1f}s</td>' \
+                f'<td>{r["vram_before_mb"]}MB</td><td>{r["vram_after_mb"]}MB</td>' \
+                f'<td>{r["vram_delta_mb"]:+d}</td><td>{status}</td></tr>'
+
+    html += '</tbody></table>'
+
+    # 图片画廊
+    saved = [r for r in results if r.get("image_path") and os.path.isfile(r["image_path"])]
+    if saved:
+        html += '<div class="section-title">生成图片</div><div class="gallery">'
+        for r in saved:
+            rel = os.path.relpath(r["image_path"], OUT_DIR)
+            html += f'<div class="item"><img src="{rel}"><div class="cap">{r["hair_style_name"]} ({r["elapsed_s"]:.1f}s)</div></div>'
+        html += '</div>'
+
+    html += '</body></html>'
+    with open(REPORT_HTML, "w", encoding="utf-8") as f:
+        f.write(html)
+
+
+if __name__ == "__main__":
+    run_test()
@@ -0,0 +1,43 @@
+nohup: ignoring input
+9B vs 4B 对比测试 — Plan=A
+图片: ['girl1.jpg', 'girl7.jpg', 'girl13.jpg']
+发型: [(1, '椭圆'), (2, '花瓣'), (3, '心形'), (4, '直线'), (5, '波浪')]
+总调用: 15
+
+[warmup] 预热请求 (girl13, style=1) ...
+  warmup: 25.7s
+
+[img1/3] girl1.jpg style=1(椭圆) ...
+  → ✓ code=0 time=9.4s vram=26207→29449MB (Δ+3242) img=saved
+[img1/3] girl1.jpg style=2(花瓣) ...
+  → ✓ code=0 time=10.5s vram=29449→26341MB (Δ-3108) img=saved
+[img1/3] girl1.jpg style=3(心形) ...
+  → ✓ code=0 time=10.5s vram=26341→26051MB (Δ-290) img=saved
+[img1/3] girl1.jpg style=4(直线) ...
+  → ✓ code=0 time=10.4s vram=26051→26257MB (Δ+206) img=saved
+[img1/3] girl1.jpg style=5(波浪) ...
+  → ✓ code=0 time=10.8s vram=26257→26319MB (Δ+62) img=saved
+[img2/3] girl7.jpg style=1(椭圆) ...
+  → ✓ code=0 time=7.9s vram=26319→28417MB (Δ+2098) img=saved
+[img2/3] girl7.jpg style=2(花瓣) ...
+  → ✓ code=0 time=7.8s vram=28417→28397MB (Δ-20) img=saved
+[img2/3] girl7.jpg style=3(心形) ...
+  → ✓ code=0 time=7.7s vram=28397→28409MB (Δ+12) img=saved
+[img2/3] girl7.jpg style=4(直线) ...
+  → ✓ code=0 time=8.2s vram=28409→26335MB (Δ-2074) img=saved
+[img2/3] girl7.jpg style=5(波浪) ...
+  → ✓ code=0 time=7.9s vram=26335→28419MB (Δ+2084) img=saved
+[img3/3] girl13.jpg style=1(椭圆) ...
+  → ✓ code=0 time=8.9s vram=28419→28835MB (Δ+416) img=saved
+[img3/3] girl13.jpg style=2(花瓣) ...
+  → ✓ code=0 time=9.1s vram=28835→28817MB (Δ-18) img=saved
+[img3/3] girl13.jpg style=3(心形) ...
+  → ✓ code=0 time=8.8s vram=28817→28795MB (Δ-22) img=saved
+[img3/3] girl13.jpg style=4(直线) ...
+  → ✓ code=0 time=8.8s vram=28795→28845MB (Δ+50) img=saved
+[img3/3] girl13.jpg style=5(波浪) ...
+  → ✓ code=0 time=9.0s vram=28845→26291MB (Δ-2554) img=saved
+
+✅ 完成！/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/results.json
+📊 成功: 15/15
+⏱  平均: 9.06s
@@ -0,0 +1,257 @@
+[
+  {
+    "plan": "A",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:04:32",
+    "elapsed_s": 9.41,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26207,
+    "vram_after_mb": 29449,
+    "vram_delta_mb": 3242,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img1_style1.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:04:42",
+    "elapsed_s": 10.51,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 29449,
+    "vram_after_mb": 26341,
+    "vram_delta_mb": -3108,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img1_style2.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:04:53",
+    "elapsed_s": 10.49,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26341,
+    "vram_after_mb": 26051,
+    "vram_delta_mb": -290,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img1_style3.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:05:03",
+    "elapsed_s": 10.45,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26051,
+    "vram_after_mb": 26257,
+    "vram_delta_mb": 206,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img1_style4.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:05:14",
+    "elapsed_s": 10.85,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26257,
+    "vram_after_mb": 26319,
+    "vram_delta_mb": 62,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img1_style5.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:05:22",
+    "elapsed_s": 7.91,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26319,
+    "vram_after_mb": 28417,
+    "vram_delta_mb": 2098,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img2_style1.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:05:30",
+    "elapsed_s": 7.81,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28417,
+    "vram_after_mb": 28397,
+    "vram_delta_mb": -20,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img2_style2.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:05:38",
+    "elapsed_s": 7.74,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28397,
+    "vram_after_mb": 28409,
+    "vram_delta_mb": 12,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img2_style3.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:05:46",
+    "elapsed_s": 8.17,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28409,
+    "vram_after_mb": 26335,
+    "vram_delta_mb": -2074,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img2_style4.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:05:54",
+    "elapsed_s": 7.92,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 26335,
+    "vram_after_mb": 28419,
+    "vram_delta_mb": 2084,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img2_style5.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:06:03",
+    "elapsed_s": 8.95,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28419,
+    "vram_after_mb": 28835,
+    "vram_delta_mb": 416,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img3_style1.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:06:12",
+    "elapsed_s": 9.07,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28835,
+    "vram_after_mb": 28817,
+    "vram_delta_mb": -18,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img3_style2.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:06:21",
+    "elapsed_s": 8.82,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28817,
+    "vram_after_mb": 28795,
+    "vram_delta_mb": -22,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img3_style3.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:06:30",
+    "elapsed_s": 8.84,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28795,
+    "vram_after_mb": 28845,
+    "vram_delta_mb": 50,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img3_style4.jpg"
+  },
+  {
+    "plan": "A",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:06:39",
+    "elapsed_s": 8.96,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28845,
+    "vram_after_mb": 26291,
+    "vram_delta_mb": -2554,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_A/img3_style5.jpg"
+  }
+]
@@ -0,0 +1,43 @@
+nohup: ignoring input
+9B vs 4B 对比测试 — Plan=Bplus
+图片: ['girl1.jpg', 'girl7.jpg', 'girl13.jpg']
+发型: [(1, '椭圆'), (2, '花瓣'), (3, '心形'), (4, '直线'), (5, '波浪')]
+总调用: 15
+
+[warmup] 预热请求 (girl13, style=1) ...
+  warmup: 12.1s
+
+[img1/3] girl1.jpg style=1(椭圆) ...
+  → ✓ code=0 time=7.7s vram=25531→28691MB (Δ+3160) img=saved
+[img1/3] girl1.jpg style=2(花瓣) ...
+  → ✓ code=0 time=8.8s vram=28691→28655MB (Δ-36) img=saved
+[img1/3] girl1.jpg style=3(心形) ...
+  → ✓ code=0 time=9.1s vram=28655→28651MB (Δ-4) img=saved
+[img1/3] girl1.jpg style=4(直线) ...
+  → ✓ code=0 time=8.7s vram=28651→28641MB (Δ-10) img=saved
+[img1/3] girl1.jpg style=5(波浪) ...
+  → ✓ code=0 time=8.8s vram=28641→28637MB (Δ-4) img=saved
+[img2/3] girl7.jpg style=1(椭圆) ...
+  → ✓ code=0 time=7.0s vram=28637→25539MB (Δ-3098) img=saved
+[img2/3] girl7.jpg style=2(花瓣) ...
+  → ✓ code=0 time=6.7s vram=25539→27599MB (Δ+2060) img=saved
+[img2/3] girl7.jpg style=3(心形) ...
+  → ✓ code=0 time=6.7s vram=27599→27597MB (Δ-2) img=saved
+[img2/3] girl7.jpg style=4(直线) ...
+  → ✓ code=0 time=7.0s vram=27597→25533MB (Δ-2064) img=saved
+[img2/3] girl7.jpg style=5(波浪) ...
+  → ✓ code=0 time=6.7s vram=25533→27585MB (Δ+2052) img=saved
+[img3/3] girl13.jpg style=1(椭圆) ...
+  → ✓ code=0 time=7.6s vram=27585→28023MB (Δ+438) img=saved
+[img3/3] girl13.jpg style=2(花瓣) ...
+  → ✓ code=0 time=7.8s vram=28023→27997MB (Δ-26) img=saved
+[img3/3] girl13.jpg style=3(心形) ...
+  → ✓ code=0 time=7.8s vram=27997→25557MB (Δ-2440) img=saved
+[img3/3] girl13.jpg style=4(直线) ...
+  → ✓ code=0 time=7.6s vram=25557→28011MB (Δ+2454) img=saved
+[img3/3] girl13.jpg style=5(波浪) ...
+  → ✓ code=0 time=7.8s vram=28011→28017MB (Δ+6) img=saved
+
+✅ 完成！/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/results.json
+📊 成功: 15/15
+⏱  平均: 7.72s
@@ -0,0 +1,257 @@
+[
+  {
+    "plan": "Bplus",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:08:20",
+    "elapsed_s": 7.69,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25531,
+    "vram_after_mb": 28691,
+    "vram_delta_mb": 3160,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img1_style1.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:08:28",
+    "elapsed_s": 8.77,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28691,
+    "vram_after_mb": 28655,
+    "vram_delta_mb": -36,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img1_style2.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:08:38",
+    "elapsed_s": 9.07,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28655,
+    "vram_after_mb": 28651,
+    "vram_delta_mb": -4,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img1_style3.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:08:46",
+    "elapsed_s": 8.73,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28651,
+    "vram_after_mb": 28641,
+    "vram_delta_mb": -10,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img1_style4.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 1,
+    "image_name": "girl1.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl1.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:08:55",
+    "elapsed_s": 8.81,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28641,
+    "vram_after_mb": 28637,
+    "vram_delta_mb": -4,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img1_style5.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:09:02",
+    "elapsed_s": 7.04,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28637,
+    "vram_after_mb": 25539,
+    "vram_delta_mb": -3098,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img2_style1.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:09:09",
+    "elapsed_s": 6.67,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25539,
+    "vram_after_mb": 27599,
+    "vram_delta_mb": 2060,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img2_style2.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:09:16",
+    "elapsed_s": 6.73,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27599,
+    "vram_after_mb": 27597,
+    "vram_delta_mb": -2,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img2_style3.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:09:23",
+    "elapsed_s": 6.96,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27597,
+    "vram_after_mb": 25533,
+    "vram_delta_mb": -2064,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img2_style4.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 2,
+    "image_name": "girl7.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl7.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:09:29",
+    "elapsed_s": 6.74,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25533,
+    "vram_after_mb": 27585,
+    "vram_delta_mb": 2052,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img2_style5.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "14:09:37",
+    "elapsed_s": 7.62,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27585,
+    "vram_after_mb": 28023,
+    "vram_delta_mb": 438,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img3_style1.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "14:09:45",
+    "elapsed_s": 7.84,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28023,
+    "vram_after_mb": 27997,
+    "vram_delta_mb": -26,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img3_style2.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "14:09:53",
+    "elapsed_s": 7.83,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27997,
+    "vram_after_mb": 25557,
+    "vram_delta_mb": -2440,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img3_style3.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "14:10:00",
+    "elapsed_s": 7.59,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25557,
+    "vram_after_mb": 28011,
+    "vram_delta_mb": 2454,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img3_style4.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "image_idx": 3,
+    "image_name": "girl13.jpg",
+    "image_path": "/home/ubuntu/hair/image/girl_img/girl13.jpg",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "14:10:08",
+    "elapsed_s": 7.77,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28011,
+    "vram_after_mb": 28017,
+    "vram_delta_mb": 6,
+    "saved_image_path": "/home/ubuntu/hair/benchmark_out/9b_vs_4b_Bplus/img3_style5.jpg"
+  }
+]
@@ -0,0 +1,292 @@
+nohup: ignoring input
+找到 19 张图片
+总计 95 次调用，已完成 0，剩余 95
+
+[1/19] girl1.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.8s vram=22353→22353MB preview=✓ grown=✓
+
+[1/19] girl1.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=13.7s vram=22353→22389MB preview=✓ grown=✓
+
+[1/19] girl1.jpg → 心形(3) ...
+  → HTTP code=0 time=13.6s vram=22389→22375MB preview=✓ grown=✓
+
+[1/19] girl1.jpg → 直线形(4) ...
+  → HTTP code=0 time=14.0s vram=22375→22359MB preview=✓ grown=✓
+
+[1/19] girl1.jpg → 波浪形(5) ...
+  → HTTP code=0 time=13.6s vram=22359→22393MB preview=✓ grown=✓
+
+[2/19] girl10.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=9.6s vram=22393→22413MB preview=✓ grown=✓
+
+[2/19] girl10.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=9.8s vram=22413→22397MB preview=✓ grown=✓
+
+[2/19] girl10.jpg → 心形(3) ...
+  → HTTP code=0 time=9.8s vram=22397→22391MB preview=✓ grown=✓
+
+[2/19] girl10.jpg → 直线形(4) ...
+  → HTTP code=0 time=9.8s vram=22391→22393MB preview=✓ grown=✓
+
+[2/19] girl10.jpg → 波浪形(5) ...
+  → HTTP code=0 time=9.8s vram=22393→22395MB preview=✓ grown=✓
+
+[3/19] girl11.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=10.0s vram=22395→22421MB preview=✓ grown=✓
+
+[3/19] girl11.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=10.3s vram=22421→22393MB preview=✓ grown=✓
+
+[3/19] girl11.jpg → 心形(3) ...
+  → HTTP code=0 time=10.4s vram=22393→22375MB preview=✓ grown=✓
+
+[3/19] girl11.jpg → 直线形(4) ...
+  → HTTP code=0 time=10.4s vram=22375→22385MB preview=✓ grown=✓
+
+[3/19] girl11.jpg → 波浪形(5) ...
+  → HTTP code=0 time=9.9s vram=22385→22389MB preview=✓ grown=✓
+
+[4/19] girl12.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=13.0s vram=22389→22389MB preview=✓ grown=✓
+
+[4/19] girl12.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=12.9s vram=22389→22385MB preview=✓ grown=✓
+
+[4/19] girl12.jpg → 心形(3) ...
+  → HTTP code=0 time=12.9s vram=22385→22389MB preview=✓ grown=✓
+
+[4/19] girl12.jpg → 直线形(4) ...
+  → HTTP code=0 time=13.0s vram=22389→22389MB preview=✓ grown=✓
+
+[4/19] girl12.jpg → 波浪形(5) ...
+  → HTTP code=0 time=13.3s vram=22389→22409MB preview=✓ grown=✓
+
+[5/19] girl13.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.5s vram=22409→22365MB preview=✓ grown=✓
+
+[5/19] girl13.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.2s vram=22365→22395MB preview=✓ grown=✓
+
+[5/19] girl13.jpg → 心形(3) ...
+  → HTTP code=0 time=11.2s vram=22395→22421MB preview=✓ grown=✓
+
+[5/19] girl13.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.4s vram=22421→22409MB preview=✓ grown=✓
+
+[5/19] girl13.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.2s vram=22409→22431MB preview=✓ grown=✓
+
+[6/19] girl14.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.5s vram=22431→22391MB preview=✓ grown=✓
+
+[6/19] girl14.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.5s vram=22391→22375MB preview=✓ grown=✓
+
+[6/19] girl14.jpg → 心形(3) ...
+  → HTTP code=0 time=11.5s vram=22375→22409MB preview=✓ grown=✓
+
+[6/19] girl14.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.5s vram=22409→22359MB preview=✓ grown=✓
+
+[6/19] girl14.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.5s vram=22359→22393MB preview=✓ grown=✓
+
+[7/19] girl15.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.2s vram=22393→22385MB preview=✓ grown=✓
+
+[7/19] girl15.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.3s vram=22385→22387MB preview=✓ grown=✓
+
+[7/19] girl15.jpg → 心形(3) ...
+  → HTTP code=0 time=11.4s vram=22387→22357MB preview=✓ grown=✓
+
+[7/19] girl15.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.3s vram=22357→22383MB preview=✓ grown=✓
+
+[7/19] girl15.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.7s vram=22383→22387MB preview=✓ grown=✓
+
+[8/19] girl16.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.2s vram=22387→22377MB preview=✓ grown=✓
+
+[8/19] girl16.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.2s vram=22377→22359MB preview=✓ grown=✓
+
+[8/19] girl16.jpg → 心形(3) ...
+  → HTTP code=0 time=11.2s vram=22359→22347MB preview=✓ grown=✓
+
+[8/19] girl16.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.3s vram=22347→22367MB preview=✓ grown=✓
+
+[8/19] girl16.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.4s vram=22367→22399MB preview=✓ grown=✓
+
+[9/19] girl17.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=13.0s vram=22399→22403MB preview=✓ grown=✓
+
+[9/19] girl17.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=13.0s vram=22403→22421MB preview=✓ grown=✓
+
+[9/19] girl17.jpg → 心形(3) ...
+  → HTTP code=0 time=13.1s vram=22421→22381MB preview=✓ grown=✓
+
+[9/19] girl17.jpg → 直线形(4) ...
+  → HTTP code=0 time=13.1s vram=22381→22419MB preview=✓ grown=✓
+
+[9/19] girl17.jpg → 波浪形(5) ...
+  → HTTP code=0 time=13.0s vram=22419→22399MB preview=✓ grown=✓
+
+[10/19] girl18.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.6s vram=22399→22419MB preview=✓ grown=✓
+
+[10/19] girl18.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.2s vram=22419→22383MB preview=✓ grown=✓
+
+[10/19] girl18.jpg → 心形(3) ...
+  → HTTP code=0 time=11.4s vram=22383→22409MB preview=✓ grown=✓
+
+[10/19] girl18.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.4s vram=22409→22375MB preview=✓ grown=✓
+
+[10/19] girl18.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.3s vram=22375→22413MB preview=✓ grown=✓
+
+[11/19] girl19.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.1s vram=22413→22385MB preview=✓ grown=✓
+
+[11/19] girl19.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.1s vram=22385→22393MB preview=✓ grown=✓
+
+[11/19] girl19.jpg → 心形(3) ...
+  → HTTP code=0 time=11.3s vram=22393→22387MB preview=✓ grown=✓
+
+[11/19] girl19.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.2s vram=22387→22415MB preview=✓ grown=✓
+
+[11/19] girl19.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.2s vram=22415→22405MB preview=✓ grown=✓
+
+[12/19] girl2.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=13.1s vram=22405→22395MB preview=✓ grown=✓
+
+[12/19] girl2.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=12.9s vram=22395→22387MB preview=✓ grown=✓
+
+[12/19] girl2.jpg → 心形(3) ...
+  → HTTP code=0 time=12.9s vram=22387→22381MB preview=✓ grown=✓
+
+[12/19] girl2.jpg → 直线形(4) ...
+  → HTTP code=0 time=12.9s vram=22381→22387MB preview=✓ grown=✓
+
+[12/19] girl2.jpg → 波浪形(5) ...
+  → HTTP code=0 time=12.9s vram=22387→22309MB preview=✓ grown=✓
+
+[13/19] girl3.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=9.8s vram=22309→22381MB preview=✓ grown=✓
+
+[13/19] girl3.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=9.8s vram=22381→22403MB preview=✓ grown=✓
+
+[13/19] girl3.jpg → 心形(3) ...
+  → HTTP code=0 time=9.9s vram=22403→22419MB preview=✓ grown=✓
+
+[13/19] girl3.jpg → 直线形(4) ...
+  → HTTP code=0 time=9.9s vram=22419→22401MB preview=✓ grown=✓
+
+[13/19] girl3.jpg → 波浪形(5) ...
+  → HTTP code=0 time=9.9s vram=22401→22385MB preview=✓ grown=✓
+
+[14/19] girl4.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=10.2s vram=22385→22377MB preview=✓ grown=✓
+
+[14/19] girl4.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=10.5s vram=22377→22387MB preview=✓ grown=✓
+
+[14/19] girl4.jpg → 心形(3) ...
+  → HTTP code=0 time=10.0s vram=22387→22365MB preview=✓ grown=✓
+
+[14/19] girl4.jpg → 直线形(4) ...
+  → HTTP code=0 time=9.9s vram=22365→22371MB preview=✓ grown=✓
+
+[14/19] girl4.jpg → 波浪形(5) ...
+  → HTTP code=0 time=10.0s vram=22371→22389MB preview=✓ grown=✓
+
+[15/19] girl5.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=8.5s vram=22389→23903MB preview=✓ grown=✓
+
+[15/19] girl5.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=8.6s vram=23903→23929MB preview=✓ grown=✓
+
+[15/19] girl5.jpg → 心形(3) ...
+  → HTTP code=0 time=8.6s vram=23929→23925MB preview=✓ grown=✓
+
+[15/19] girl5.jpg → 直线形(4) ...
+  → HTTP code=0 time=8.7s vram=23925→23931MB preview=✓ grown=✓
+
+[15/19] girl5.jpg → 波浪形(5) ...
+  → HTTP code=0 time=8.7s vram=23931→23951MB preview=✓ grown=✓
+
+[16/19] girl6.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=8.4s vram=23951→23729MB preview=✓ grown=✓
+
+[16/19] girl6.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=8.7s vram=23729→22377MB preview=✓ grown=✓
+
+[16/19] girl6.jpg → 心形(3) ...
+  → HTTP code=0 time=8.4s vram=22377→23685MB preview=✓ grown=✓
+
+[16/19] girl6.jpg → 直线形(4) ...
+  → HTTP code=0 time=8.3s vram=23685→23731MB preview=✓ grown=✓
+
+[16/19] girl6.jpg → 波浪形(5) ...
+  → HTTP code=0 time=8.4s vram=23731→23713MB preview=✓ grown=✓
+
+[17/19] girl7.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=9.9s vram=23713→24481MB preview=✓ grown=✓
+
+[17/19] girl7.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=10.1s vram=24481→24447MB preview=✓ grown=✓
+
+[17/19] girl7.jpg → 心形(3) ...
+  → HTTP code=0 time=10.0s vram=24447→24481MB preview=✓ grown=✓
+
+[17/19] girl7.jpg → 直线形(4) ...
+  → HTTP code=0 time=10.0s vram=24481→24469MB preview=✓ grown=✓
+
+[17/19] girl7.jpg → 波浪形(5) ...
+  → HTTP code=0 time=10.0s vram=24469→24475MB preview=✓ grown=✓
+
+[18/19] girl8.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.4s vram=24475→24861MB preview=✓ grown=✓
+
+[18/19] girl8.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.5s vram=24861→24823MB preview=✓ grown=✓
+
+[18/19] girl8.jpg → 心形(3) ...
+  → HTTP code=0 time=11.3s vram=24823→22393MB preview=✓ grown=✓
+
+[18/19] girl8.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.1s vram=22393→22389MB preview=✓ grown=✓
+
+[18/19] girl8.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.1s vram=22389→22365MB preview=✓ grown=✓
+
+[19/19] girl9.jpg → 椭圆形(1) ...
+  → HTTP code=0 time=11.9s vram=22365→22391MB preview=✓ grown=✓
+
+[19/19] girl9.jpg → 花瓣形(2) ...
+  → HTTP code=0 time=11.9s vram=22391→22375MB preview=✓ grown=✓
+
+[19/19] girl9.jpg → 心形(3) ...
+  → HTTP code=0 time=11.9s vram=22375→22391MB preview=✓ grown=✓
+
+[19/19] girl9.jpg → 直线形(4) ...
+  → HTTP code=0 time=11.9s vram=22391→22371MB preview=✓ grown=✓
+
+[19/19] girl9.jpg → 波浪形(5) ...
+  → HTTP code=0 time=11.8s vram=22371→22373MB preview=✓ grown=✓
+
+✅ 完成！结果: /home/ubuntu/hair/benchmark_out/benchmark_results.json
+📄 报告: /home/ubuntu/hair/benchmark_out/benchmark_report.html
+📊 总调用: 95/95
@@ -0,0 +1,22 @@
+nohup: ignoring input
+接口2女性5种发型测试 — Plan=B
+图片：/home/ubuntu/hair/image/girl_img/girl13.jpg
+
+
+[1/5] hair_style=1 (椭圆) ...
+  → ✓ code=0 time=8.7s vram=24981→25025MB (Δ+44) img=saved
+
+[2/5] hair_style=2 (花瓣) ...
+  → ✓ code=0 time=8.4s vram=25025→27451MB (Δ+2426) img=saved
+
+[3/5] hair_style=3 (心形) ...
+  → ✓ code=0 time=8.4s vram=27451→27431MB (Δ-20) img=saved
+
+[4/5] hair_style=4 (直线) ...
+  → ✓ code=0 time=8.5s vram=27431→27433MB (Δ+2) img=saved
+
+[5/5] hair_style=5 (波浪) ...
+  → ✓ code=0 time=8.4s vram=27433→27407MB (Δ-26) img=saved
+
+✅ 完成！/home/ubuntu/hair/benchmark_out/iface2_female_B/results.json
+📄 报告：/home/ubuntu/hair/benchmark_out/iface2_female_B/report.html
@@ -0,0 +1,34 @@
+<!DOCTYPE html>
+<html lang="zh-CN"><head><meta charset="UTF-8"><title>接口2女5种发型 - Plan B</title>
+<style>
+*{margin:0;padding:0;box-sizing:border-box}
+body{font-family:sans-serif;background:#1a1a2e;color:#e0e0e0;padding:20px}
+h1{text-align:center;color:#00d4ff;margin-bottom:10px}
+.subtitle{text-align:center;color:#888;margin-bottom:30px;font-size:14px}
+.summary{display:grid;grid-template-columns:repeat(auto-fit,minmax(160px,1fr));gap:15px;margin-bottom:30px}
+.card{background:#16213e;border-radius:12px;padding:20px;text-align:center;border:1px solid #333}
+.card .num{font-size:28px;font-weight:700}
+.card .label{font-size:12px;color:#888;margin-top:5px}
+.card.ok .num{color:#0f0} .card.err .num{color:#f44} .card.time .num{color:#00d4ff}
+table{width:100%;border-collapse:collapse;margin-bottom:30px;background:#16213e;border-radius:12px;overflow:hidden}
+th{background:#0f3460;padding:10px 8px;text-align:center;font-size:13px;color:#fff}
+td{padding:8px;text-align:center;border-bottom:1px solid #222;font-size:13px}
+tr:hover{background:#1a1a3e}
+.section-title{font-size:18px;font-weight:600;margin:30px 0 15px;color:#00d4ff;border-bottom:1px solid #333;padding-bottom:10px}
+.ok{color:#0f0} .fail{color:#f44}
+.time-bar{display:inline-block;height:18px;background:linear-gradient(90deg,#0f3460,#00d4ff);border-radius:3px;vertical-align:middle;min-width:2px}
+.gallery{display:grid;grid-template-columns:repeat(auto-fit,minmax(200px,1fr));gap:15px}
+.gallery img{width:100%;border-radius:8px;border:1px solid #333}
+.gallery .item{text-align:center}
+.gallery .cap{margin-top:5px;font-size:12px;color:#888}
+</style></head><body>
+<h1>📊 接口2女性5种发型测试报告</h1>
+<div class="subtitle">Plan B ｜ 2026-07-18 13:52:20 ｜ 5 次调用</div>
+<div class="summary">
+<div class="card ok"><div class="num">5</div><div class="label">成功</div></div>
+<div class="card err"><div class="num">0</div><div class="label">失败</div></div>
+<div class="card time"><div class="num">8.5s</div><div class="label">平均耗时</div></div>
+</div>
+<div class="section-title">详细结果</div>
+<table><thead><tr><th>发型</th><th>名称</th><th>耗时</th><th>显存前</th><th>显存后</th><th>Δ</th><th>状态</th></tr></thead><tbody>
+<tr><td>style 1</td><td>椭圆</td><td><div class="time-bar" style="width:130.65px"></div> 8.7s</td><td>24981MB</td><td>25025MB</td><td>+44</td><td><span class="ok">✓</span></td></tr><tr><td>style 2</td><td>花瓣</td><td><div class="time-bar" style="width:125.54999999999998px"></div> 8.4s</td><td>25025MB</td><td>27451MB</td><td>+2426</td><td><span class="ok">✓</span></td></tr><tr><td>style 3</td><td>心形</td><td><div class="time-bar" style="width:125.54999999999998px"></div> 8.4s</td><td>27451MB</td><td>27431MB</td><td>-20</td><td><span class="ok">✓</span></td></tr><tr><td>style 4</td><td>直线</td><td><div class="time-bar" style="width:127.5px"></div> 8.5s</td><td>27431MB</td><td>27433MB</td><td>+2</td><td><span class="ok">✓</span></td></tr><tr><td>style 5</td><td>波浪</td><td><div class="time-bar" style="width:126.0px"></div> 8.4s</td><td>27433MB</td><td>27407MB</td><td>-26</td><td><span class="ok">✓</span></td></tr></tbody></table><div class="section-title">生成图片</div><div class="gallery"><div class="item"><img src="hair_style_1_椭圆.jpg"><div class="cap">椭圆 (8.7s)</div></div><div class="item"><img src="hair_style_2_花瓣.jpg"><div class="cap">花瓣 (8.4s)</div></div><div class="item"><img src="hair_style_3_心形.jpg"><div class="cap">心形 (8.4s)</div></div><div class="item"><img src="hair_style_4_直线.jpg"><div class="cap">直线 (8.5s)</div></div><div class="item"><img src="hair_style_5_波浪.jpg"><div class="cap">波浪 (8.4s)</div></div></div></body></html>
@@ -0,0 +1,72 @@
+[
+  {
+    "plan": "B",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "13:51:46",
+    "elapsed_s": 8.71,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 24981,
+    "vram_after_mb": 25025,
+    "vram_delta_mb": 44,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_B/hair_style_1_椭圆.jpg"
+  },
+  {
+    "plan": "B",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "13:51:55",
+    "elapsed_s": 8.37,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25025,
+    "vram_after_mb": 27451,
+    "vram_delta_mb": 2426,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_B/hair_style_2_花瓣.jpg"
+  },
+  {
+    "plan": "B",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "13:52:03",
+    "elapsed_s": 8.37,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27451,
+    "vram_after_mb": 27431,
+    "vram_delta_mb": -20,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_B/hair_style_3_心形.jpg"
+  },
+  {
+    "plan": "B",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "13:52:11",
+    "elapsed_s": 8.5,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27431,
+    "vram_after_mb": 27433,
+    "vram_delta_mb": 2,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_B/hair_style_4_直线.jpg"
+  },
+  {
+    "plan": "B",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "13:52:20",
+    "elapsed_s": 8.4,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 27433,
+    "vram_after_mb": 27407,
+    "vram_delta_mb": -26,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_B/hair_style_5_波浪.jpg"
+  }
+]
@@ -0,0 +1,22 @@
+nohup: ignoring input
+接口2女性5种发型测试 — Plan=Bplus
+图片：/home/ubuntu/hair/image/girl_img/girl13.jpg
+
+
+[1/5] hair_style=1 (椭圆) ...
+  → ✓ code=0 time=7.9s vram=25875→25889MB (Δ+14) img=saved
+
+[2/5] hair_style=2 (花瓣) ...
+  → ✓ code=0 time=8.0s vram=25889→28337MB (Δ+2448) img=saved
+
+[3/5] hair_style=3 (心形) ...
+  → ✓ code=0 time=7.8s vram=28337→28355MB (Δ+18) img=saved
+
+[4/5] hair_style=4 (直线) ...
+  → ✓ code=0 time=7.6s vram=28355→28333MB (Δ-22) img=saved
+
+[5/5] hair_style=5 (波浪) ...
+  → ✓ code=0 time=8.3s vram=28333→25885MB (Δ-2448) img=saved
+
+✅ 完成！/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/results.json
+📄 报告：/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/report.html
@@ -0,0 +1,34 @@
+<!DOCTYPE html>
+<html lang="zh-CN"><head><meta charset="UTF-8"><title>接口2女5种发型 - Plan Bplus</title>
+<style>
+*{margin:0;padding:0;box-sizing:border-box}
+body{font-family:sans-serif;background:#1a1a2e;color:#e0e0e0;padding:20px}
+h1{text-align:center;color:#00d4ff;margin-bottom:10px}
+.subtitle{text-align:center;color:#888;margin-bottom:30px;font-size:14px}
+.summary{display:grid;grid-template-columns:repeat(auto-fit,minmax(160px,1fr));gap:15px;margin-bottom:30px}
+.card{background:#16213e;border-radius:12px;padding:20px;text-align:center;border:1px solid #333}
+.card .num{font-size:28px;font-weight:700}
+.card .label{font-size:12px;color:#888;margin-top:5px}
+.card.ok .num{color:#0f0} .card.err .num{color:#f44} .card.time .num{color:#00d4ff}
+table{width:100%;border-collapse:collapse;margin-bottom:30px;background:#16213e;border-radius:12px;overflow:hidden}
+th{background:#0f3460;padding:10px 8px;text-align:center;font-size:13px;color:#fff}
+td{padding:8px;text-align:center;border-bottom:1px solid #222;font-size:13px}
+tr:hover{background:#1a1a3e}
+.section-title{font-size:18px;font-weight:600;margin:30px 0 15px;color:#00d4ff;border-bottom:1px solid #333;padding-bottom:10px}
+.ok{color:#0f0} .fail{color:#f44}
+.time-bar{display:inline-block;height:18px;background:linear-gradient(90deg,#0f3460,#00d4ff);border-radius:3px;vertical-align:middle;min-width:2px}
+.gallery{display:grid;grid-template-columns:repeat(auto-fit,minmax(200px,1fr));gap:15px}
+.gallery img{width:100%;border-radius:8px;border:1px solid #333}
+.gallery .item{text-align:center}
+.gallery .cap{margin-top:5px;font-size:12px;color:#888}
+</style></head><body>
+<h1>📊 接口2女性5种发型测试报告</h1>
+<div class="subtitle">Plan Bplus ｜ 2026-07-18 13:50:43 ｜ 5 次调用</div>
+<div class="summary">
+<div class="card ok"><div class="num">5</div><div class="label">成功</div></div>
+<div class="card err"><div class="num">0</div><div class="label">失败</div></div>
+<div class="card time"><div class="num">7.9s</div><div class="label">平均耗时</div></div>
+</div>
+<div class="section-title">详细结果</div>
+<table><thead><tr><th>发型</th><th>名称</th><th>耗时</th><th>显存前</th><th>显存后</th><th>Δ</th><th>状态</th></tr></thead><tbody>
+<tr><td>style 1</td><td>椭圆</td><td><div class="time-bar" style="width:119.10000000000001px"></div> 7.9s</td><td>25875MB</td><td>25889MB</td><td>+14</td><td><span class="ok">✓</span></td></tr><tr><td>style 2</td><td>花瓣</td><td><div class="time-bar" style="width:120.6px"></div> 8.0s</td><td>25889MB</td><td>28337MB</td><td>+2448</td><td><span class="ok">✓</span></td></tr><tr><td>style 3</td><td>心形</td><td><div class="time-bar" style="width:116.55px"></div> 7.8s</td><td>28337MB</td><td>28355MB</td><td>+18</td><td><span class="ok">✓</span></td></tr><tr><td>style 4</td><td>直线</td><td><div class="time-bar" style="width:114.6px"></div> 7.6s</td><td>28355MB</td><td>28333MB</td><td>-22</td><td><span class="ok">✓</span></td></tr><tr><td>style 5</td><td>波浪</td><td><div class="time-bar" style="width:123.89999999999999px"></div> 8.3s</td><td>28333MB</td><td>25885MB</td><td>-2448</td><td><span class="ok">✓</span></td></tr></tbody></table><div class="section-title">生成图片</div><div class="gallery"><div class="item"><img src="hair_style_1_椭圆.jpg"><div class="cap">椭圆 (7.9s)</div></div><div class="item"><img src="hair_style_2_花瓣.jpg"><div class="cap">花瓣 (8.0s)</div></div><div class="item"><img src="hair_style_3_心形.jpg"><div class="cap">心形 (7.8s)</div></div><div class="item"><img src="hair_style_4_直线.jpg"><div class="cap">直线 (7.6s)</div></div><div class="item"><img src="hair_style_5_波浪.jpg"><div class="cap">波浪 (8.3s)</div></div></div></body></html>
@@ -0,0 +1,72 @@
+[
+  {
+    "plan": "Bplus",
+    "hair_style": 1,
+    "hair_style_name": "椭圆",
+    "timestamp": "13:50:11",
+    "elapsed_s": 7.94,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25875,
+    "vram_after_mb": 25889,
+    "vram_delta_mb": 14,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/hair_style_1_椭圆.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "hair_style": 2,
+    "hair_style_name": "花瓣",
+    "timestamp": "13:50:19",
+    "elapsed_s": 8.04,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25889,
+    "vram_after_mb": 28337,
+    "vram_delta_mb": 2448,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/hair_style_2_花瓣.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "hair_style": 3,
+    "hair_style_name": "心形",
+    "timestamp": "13:50:27",
+    "elapsed_s": 7.77,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28337,
+    "vram_after_mb": 28355,
+    "vram_delta_mb": 18,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/hair_style_3_心形.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "hair_style": 4,
+    "hair_style_name": "直线",
+    "timestamp": "13:50:35",
+    "elapsed_s": 7.64,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28355,
+    "vram_after_mb": 28333,
+    "vram_delta_mb": -22,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/hair_style_4_直线.jpg"
+  },
+  {
+    "plan": "Bplus",
+    "hair_style": 5,
+    "hair_style_name": "波浪",
+    "timestamp": "13:50:43",
+    "elapsed_s": 8.26,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28333,
+    "vram_after_mb": 25885,
+    "vram_delta_mb": -2448,
+    "image_path": "/home/ubuntu/hair/benchmark_out/iface2_female_Bplus/hair_style_5_波浪.jpg"
+  }
+]
@@ -0,0 +1,562 @@
+[
+  {
+    "round": 1,
+    "test_id": "iface1",
+    "test_name": "接口1-测量",
+    "timestamp": "13:36:27",
+    "elapsed_s": 0.37,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25303,
+    "vram_after_mb": 25425,
+    "vram_free_before_mb": 6818,
+    "vram_free_after_mb": 6696,
+    "vram_delta_mb": 122,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 12524,
+      "298802": 1015
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 12524,
+      "298802": 1137
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 1,
+    "test_id": "iface2f",
+    "test_name": "接口2-女-椭圆",
+    "timestamp": "13:36:35",
+    "elapsed_s": 7.54,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25425,
+    "vram_after_mb": 25887,
+    "vram_free_before_mb": 6696,
+    "vram_free_after_mb": 6234,
+    "vram_delta_mb": 462,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 12524,
+      "298802": 1137
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 1,
+    "test_id": "iface2m",
+    "test_name": "接口2-男-椭圆",
+    "timestamp": "13:36:39",
+    "elapsed_s": 4.11,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25887,
+    "vram_after_mb": 28383,
+    "vram_free_before_mb": 6234,
+    "vram_free_after_mb": 3738,
+    "vram_delta_mb": 2496,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 1,
+    "test_id": "iface3",
+    "test_name": "接口3-B端生发",
+    "timestamp": "13:36:42",
+    "elapsed_s": 3.12,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28383,
+    "vram_after_mb": 28415,
+    "vram_free_before_mb": 3738,
+    "vram_free_after_mb": 3706,
+    "vram_delta_mb": 32,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 15020,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 2,
+    "test_id": "iface1",
+    "test_name": "接口1-测量",
+    "timestamp": "13:36:42",
+    "elapsed_s": 0.08,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28415,
+    "vram_after_mb": 28415,
+    "vram_free_before_mb": 3706,
+    "vram_free_after_mb": 3706,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 15020,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 15020,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 2,
+    "test_id": "iface2f",
+    "test_name": "接口2-女-椭圆",
+    "timestamp": "13:36:50",
+    "elapsed_s": 7.51,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28415,
+    "vram_after_mb": 28353,
+    "vram_free_before_mb": 3706,
+    "vram_free_after_mb": 3768,
+    "vram_delta_mb": -62,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8522,
+      "271640": 15020,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 2,
+    "test_id": "iface2m",
+    "test_name": "接口2-男-椭圆",
+    "timestamp": "13:36:54",
+    "elapsed_s": 4.0,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28353,
+    "vram_after_mb": 28353,
+    "vram_free_before_mb": 3768,
+    "vram_free_after_mb": 3768,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 2,
+    "test_id": "iface3",
+    "test_name": "接口3-B端生发",
+    "timestamp": "13:36:57",
+    "elapsed_s": 3.27,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28353,
+    "vram_after_mb": 28353,
+    "vram_free_before_mb": 3768,
+    "vram_free_after_mb": 3768,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 3,
+    "test_id": "iface1",
+    "test_name": "接口1-测量",
+    "timestamp": "13:36:58",
+    "elapsed_s": 0.07,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28353,
+    "vram_after_mb": 28353,
+    "vram_free_before_mb": 3768,
+    "vram_free_after_mb": 3768,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 3,
+    "test_id": "iface2f",
+    "test_name": "接口2-女-椭圆",
+    "timestamp": "13:37:05",
+    "elapsed_s": 7.63,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28353,
+    "vram_after_mb": 25855,
+    "vram_free_before_mb": 3768,
+    "vram_free_after_mb": 6266,
+    "vram_delta_mb": -2498,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8492,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 3,
+    "test_id": "iface2m",
+    "test_name": "接口2-男-椭圆",
+    "timestamp": "13:37:09",
+    "elapsed_s": 4.1,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25855,
+    "vram_after_mb": 28351,
+    "vram_free_before_mb": 6266,
+    "vram_free_after_mb": 3770,
+    "vram_delta_mb": 2496,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 3,
+    "test_id": "iface3",
+    "test_name": "接口3-B端生发",
+    "timestamp": "13:37:13",
+    "elapsed_s": 3.13,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28351,
+    "vram_after_mb": 28351,
+    "vram_free_before_mb": 3770,
+    "vram_free_after_mb": 3770,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 4,
+    "test_id": "iface1",
+    "test_name": "接口1-测量",
+    "timestamp": "13:37:13",
+    "elapsed_s": 0.07,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28351,
+    "vram_after_mb": 28351,
+    "vram_free_before_mb": 3770,
+    "vram_free_after_mb": 3770,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 4,
+    "test_id": "iface2f",
+    "test_name": "接口2-女-椭圆",
+    "timestamp": "13:37:20",
+    "elapsed_s": 7.32,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28351,
+    "vram_after_mb": 28337,
+    "vram_free_before_mb": 3770,
+    "vram_free_after_mb": 3784,
+    "vram_delta_mb": -14,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 4,
+    "test_id": "iface2m",
+    "test_name": "接口2-男-椭圆",
+    "timestamp": "13:37:24",
+    "elapsed_s": 4.21,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28337,
+    "vram_after_mb": 28337,
+    "vram_free_before_mb": 3784,
+    "vram_free_after_mb": 3784,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 4,
+    "test_id": "iface3",
+    "test_name": "接口3-B端生发",
+    "timestamp": "13:37:28",
+    "elapsed_s": 3.18,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28337,
+    "vram_after_mb": 28337,
+    "vram_free_before_mb": 3784,
+    "vram_free_after_mb": 3784,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 5,
+    "test_id": "iface1",
+    "test_name": "接口1-测量",
+    "timestamp": "13:37:28",
+    "elapsed_s": 0.08,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28337,
+    "vram_after_mb": 28337,
+    "vram_free_before_mb": 3784,
+    "vram_free_after_mb": 3784,
+    "vram_delta_mb": 0,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": false
+  },
+  {
+    "round": 5,
+    "test_id": "iface2f",
+    "test_name": "接口2-女-椭圆",
+    "timestamp": "13:37:35",
+    "elapsed_s": 7.43,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28337,
+    "vram_after_mb": 25855,
+    "vram_free_before_mb": 3784,
+    "vram_free_after_mb": 6266,
+    "vram_delta_mb": -2482,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8476,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 5,
+    "test_id": "iface2m",
+    "test_name": "接口2-男-椭圆",
+    "timestamp": "13:37:39",
+    "elapsed_s": 4.07,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 25855,
+    "vram_after_mb": 28351,
+    "vram_free_before_mb": 6266,
+    "vram_free_after_mb": 3770,
+    "vram_delta_mb": 2496,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 12492,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_changed": true
+  },
+  {
+    "round": 5,
+    "test_id": "iface3",
+    "test_name": "接口3-B端生发",
+    "timestamp": "13:37:43",
+    "elapsed_s": 3.11,
+    "success": true,
+    "code": 0,
+    "error": null,
+    "vram_before_mb": 28351,
+    "vram_after_mb": 28383,
+    "vram_free_before_mb": 3770,
+    "vram_free_after_mb": 3738,
+    "vram_delta_mb": 32,
+    "procs_before": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 14988,
+      "298802": 1585
+    },
+    "procs_after": {
+      "858": 3242,
+      "24879": 8490,
+      "271640": 15020,
+      "298802": 1585
+    },
+    "procs_changed": true
+  }
+]
@@ -0,0 +1,87 @@
+nohup: ignoring input
+轮换测试：4 个接口 × 5 轮 = 20 次调用
+测试图片：/home/ubuntu/hair/image/girl_img/girl13.jpg
+
+
+============================================================
+第 1/5 轮
+============================================================
+
+[1/5] 接口1-测量 ...
+  → ✓ code=0 time=0.4s vram=25303→25425MB (Δ+122) ⚠模型换出!
+
+[1/5] 接口2-女-椭圆 ...
+  → ✓ code=0 time=7.5s vram=25425→25887MB (Δ+462) ⚠模型换出!
+
+[1/5] 接口2-男-椭圆 ...
+  → ✓ code=0 time=4.1s vram=25887→28383MB (Δ+2496) ⚠模型换出!
+
+[1/5] 接口3-B端生发 ...
+  → ✓ code=0 time=3.1s vram=28383→28415MB (Δ+32) ⚠模型换出!
+
+============================================================
+第 2/5 轮
+============================================================
+
+[2/5] 接口1-测量 ...
+  → ✓ code=0 time=0.1s vram=28415→28415MB (Δ+0)
+
+[2/5] 接口2-女-椭圆 ...
+  → ✓ code=0 time=7.5s vram=28415→28353MB (Δ-62) ⚠模型换出!
+
+[2/5] 接口2-男-椭圆 ...
+  → ✓ code=0 time=4.0s vram=28353→28353MB (Δ+0)
+
+[2/5] 接口3-B端生发 ...
+  → ✓ code=0 time=3.3s vram=28353→28353MB (Δ+0)
+
+============================================================
+第 3/5 轮
+============================================================
+
+[3/5] 接口1-测量 ...
+  → ✓ code=0 time=0.1s vram=28353→28353MB (Δ+0)
+
+[3/5] 接口2-女-椭圆 ...
+  → ✓ code=0 time=7.6s vram=28353→25855MB (Δ-2498) ⚠模型换出!
+
+[3/5] 接口2-男-椭圆 ...
+  → ✓ code=0 time=4.1s vram=25855→28351MB (Δ+2496) ⚠模型换出!
+
+[3/5] 接口3-B端生发 ...
+  → ✓ code=0 time=3.1s vram=28351→28351MB (Δ+0)
+
+============================================================
+第 4/5 轮
+============================================================
+
+[4/5] 接口1-测量 ...
+  → ✓ code=0 time=0.1s vram=28351→28351MB (Δ+0)
+
+[4/5] 接口2-女-椭圆 ...
+  → ✓ code=0 time=7.3s vram=28351→28337MB (Δ-14) ⚠模型换出!
+
+[4/5] 接口2-男-椭圆 ...
+  → ✓ code=0 time=4.2s vram=28337→28337MB (Δ+0)
+
+[4/5] 接口3-B端生发 ...
+  → ✓ code=0 time=3.2s vram=28337→28337MB (Δ+0)
+
+============================================================
+第 5/5 轮
+============================================================
+
+[5/5] 接口1-测量 ...
+  → ✓ code=0 time=0.1s vram=28337→28337MB (Δ+0)
+
+[5/5] 接口2-女-椭圆 ...
+  → ✓ code=0 time=7.4s vram=28337→25855MB (Δ-2482) ⚠模型换出!
+
+[5/5] 接口2-男-椭圆 ...
+  → ✓ code=0 time=4.1s vram=25855→28351MB (Δ+2496) ⚠模型换出!
+
+[5/5] 接口3-B端生发 ...
+  → ✓ code=0 time=3.1s vram=28351→28383MB (Δ+32) ⚠模型换出!
+
+✅ 完成！/home/ubuntu/hair/benchmark_out/rotation/rotation_results.json
+📄 报告：/home/ubuntu/hair/benchmark_out/rotation/rotation_report.html
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
xslandCursor	9fb5b486c0	fix(接口4): worker 移除脸型 Mock + face_shape 改本机 MediaPipe 计算 - worker /api/v1/face/features 不再返回假成功数据，直接告知仅网关实现，避免本机误打 :8187 被 Mock 结果误导。 - 网关 ark_api_key 加载优先级改为 gateway/config.json 优先（原先误读 worker_config.json 里的失效 key）。 - 接口4 face_shape 不再采信豆包结果，改用本机 face/face_shape_classifier.py （MediaPipe 7 类）计算覆盖；其余 5 项特征仍走豆包。 - 修复 face_shape_classifier 共享 FaceMesh 实例的线程安全问题（加锁），避免网关侧接口4 并发请求时崩溃/结果错乱。 - 新增 /api/v1/debug/face-shape 调试接口 + static/test_face_shape.html 单图调试页（worker 侧）。 - 更新文档：网关机现在也需要 mediapipe/opencv-python/numpy<2。 ⚠️ 部署前提醒：网关机需先安装 mediapipe==0.10.14 / opencv-python==4.10.0.84 / numpy==1.26.4，否则接口4 会返回 1007「分析服务异常」。 Co-authored-by: Cursor <cursoragent@cursor.com>	2026-07-29 14:41:49 +08:00
xsl	a748dfd1e5	Merge branch 'main' of http://git.xiangsilian.com:3000/xsl/hair	2026-07-29 13:26:44 +08:00
xslandCursor	587876f8a0	feat: 脸型分类器(7分类)+批量报告生成基于 MediaPipe 468 关键点提取几何特征，转 z 分数后与各脸型原型加权匹配。参考分布与原型靶心取自 1093 张测试集的实测画像，不再靠人工设定绝对阈值。方形脸占比从 29.6% 降到 13.8%，两处原因：一是参考统计量原先只由 50 张样本估得，相对全量人群有系统性偏移，且三项偏移都在给方形脸加分；二是原型把 aspect_ratio 当作方形脸的主特征，但实测方脸组该值中位仅 +0.16，真正"宽"的是圆脸(+1.01)，等于在拿脸宽找方脸。原型参数在「6 张基准标注图判定不变、且领先第二名 >=3 分」的约束下搜索得到。余量约束是必要的：早前一版余量仅 0.008 分，权重写码时四舍五入就会翻转结论。测试素材(人像照片)与报告输出体积大，一并加入 .gitignore。 Co-authored-by: Cursor <cursoragent@cursor.com>	2026-07-29 13:16:43 +08:00
xsl	81c7a0b55c	feat: 接口1/5/6 发际线弃用逻辑(顶庭<0.7cm) 发际线离头顶<0.7cm时判定分割不可靠，弃用发际线：顶/上庭字段置null、face_total只算中下庭、标注图保留头顶线去掉发际线、只标中/下庭。eye1/7竖向范围改用眉心。	2026-07-27 23:21:01 +08:00
xsl	218818bdd0	docs: 同步 integration.html / 接口文档.md / test_interface5.html - 接口5: 补 generate_grow_image 参数说明(接口文档/integration/test_interface5 加控件) - 接口1/5/6: 补 left_position/right_position 字段(MediaPipe 21/251号点) - 接口4: features 字段纠正为固定6个英文字段(原误写~42项含中文, 与代码不符) - 接口7: 完全移除(代码已 deprecated=True 固定返回错误, 文档却当正常接口详述) - 错误码: 删错误的'1004已废弃'(1004仍用于接口2/5 gender校验), 补 1004 正确描述 + 1009(X-Internal-Token鉴权) - test_interface5.html: 加 generate_grow_image 复选框	2026-07-24 00:52:14 +08:00
xsl	ae47472a63	feat(接口1/5/6): 返回数据新增 left_position/right_position(MediaPipe 21/251号点) - face_mesh_landmarks.py: 加常量 LEFT_POSITION=21 / RIGHT_POSITION=251 - measure.py: MeasureResult 收 landmarks/宽高, to_response 顶层输出两点(原图像素 {x,y}, 与 landmarks 同格式) - measure_face 透传 landmarks(签名不变, 6处调用零改动); __init__ 用 None 默认值守卫向后兼容 - 三接口自动生效: 接口1/6 在 data 顶层, 接口5 在 face_measure 对象里(复用同一 to_response) - 实测坐标左右镜像合理, 44 个现有测试全过无回归	2026-07-24 00:52:14 +08:00
xsl	cd8985e93d	feat(接口5): 新增 generate_grow_image 参数控制是否生成生发效果图 - app.py: 接口5 路由加表单参数 generate_grow_image(bool, 默认 True)并透传 - hairline/service.py: generate_hairline_pngs 加同名参数, False 时跳过 ComfyUI 生发、grown_png 恒 None - 默认行为不变(向后兼容); false 时仅返回三档发际线叠图与中心点, 大幅降低耗时 - 网关字节级透传 multipart, 新参数自动到达 worker, 无需改网关	2026-07-24 00:52:04 +08:00
UbuntuandClaude	bb9f55e93c	feat(gateway): 全局串行化(并发=1) + 记录入参/出参/耗时日志并发模型从「每worker并发1 + 多worker并行」改为全局串行：同一时间只处理1个请求，其余排队；多 worker 仅作热备（主 worker 坏了才用备机）。接口4(face/features) 走豆包、不占 GPU，不纳入串行。 - pool: asyncio.Semaphore(max_global_concurrency=1) + acquire/release_global_slot （依赖单进程 uvicorn 部署，已在注释中标注） - forward: proxy_request 最外层 acquire 全局槽、try/finally 全路径释放；入参 multipart 解析挂 request.state；原重试/故障转移逻辑抽到 _dispatch_with_retries - reqlog(新): 标量入参保留；图片(file/base64)存盘转URL，绝不内嵌base64；出参递归摘要截断(landmarks/长串/大数组) - logging_middleware: RequestLogEntry 加 request_params/response_data， jsonl 全量记录；_load_from_logfile 同步映射防重启丢字段；get_stats recent 暴露 - /gateway-health 暴露 global_max/global_busy/global_waiting - config: dispatch 新增 max_global_concurrency / max_queue_wait_seconds - tests: test_reqlog + test_gateway_serialization（9 用例）顺带提交此前未提交的网关统计(daily stats 接口)与耗时看板(api_timing_dashboard.html)。 Co-Authored-By: Claude <noreply@anthropic.com>	2026-07-23 22:56:48 +08:00
xslandCursor	98b9108837	perf(接口2): 稳定混跑耗时至12s内 —— ComfyUI插队 + CLIP挪CPU + 提示词全局统一问题：接口2 与接口3/5 乱序调用时耗时抖动（最差 15~22s）。两个根因： 1. GPU 24G 常驻 21.4G，Flux-2(3.9G) 无法完全驻留显存，每次采样动态换页，速度随空闲显存波动（2s~8s）； 2. ComfyUI 单队列 FIFO，接口2 排在接口3/5 批量任务后面。改动： - hairline/comfyui.py: run() 新增 front 参数，/prompt 带 "front": true 插队到队列最前； redraw.py 透传；service.py 接口2 三处调用（女重绘 + 男有/无遮罩）传 front=True，接口3/5 仍走普通队列。 - add_hair.json / 0716add-hair-api.json: 节点61 CLIPLoader device default→cpu。 qwen CLIP(4G) 不再占显存（文本条件缓存常年命中），ComfyUI 显存 8.8G→4.5G， Flux-2 完全驻留，采样稳定 ~3-5s。代价：换 prompt 后首次请求 CPU 编码 ~11s（一次性）。 - 提示词全局统一为「填充遮罩区域的头发，皮肤加一点磨皮，再加一点美颜」： app.py 4处默认值、service.py _REDRAW_PROMPT、redraw.py _DEFAULT_PROMPT、 4个工作流节点60内置文案、测试页(test_interface2/3/7/12/12_final)、local_test。任何两个不同 prompt 交替提交都会打爆 CLIP 编码缓存（--cache-classic 只存最近一次），之前测试页旧文案与服务端不一致导致交替测试每次 +11s。 - app.py: 接口7 /api/v1/hair/grow-v2 下线（业务弃用；add_hair2.json 的 Klein-9b 会把常驻 Klein-4b 挤出显存）。保留 stub 返回 1007 明确报错，避免裸 404。实测（1024 档）：接口2女 8.5~10s、接口2男 ~5s、接口3 ~7-10s，交替混跑无尖刺。 Co-authored-by: Cursor <cursoragent@cursor.com> (cherry-picked from ubuntu3090 e7b62f2；已适配 main 分支代码结构：main 无 _REDRAW_PROMPT/_REDRAW_MAX_SIDE 缩图逻辑，front=True 直接加在 _call_local_redraw / generate_grow_results 的调用点；另把 main 独有的 benchmark_*.py 里的 prompt 一并统一) Co-authored-by: Cursor <cursoragent@cursor.com>	2026-07-23 01:35:42 +08:00
xslandCursor	d4aae9722c	fix(pose): 修正正面照被误判为1003（solvePnP翻转解） ITERATIVE 偶发收敛到相机后方(tz<0)，roll≈±180° 超阈值，把正面照误判为非正面。检测到负深度时回退 SQPNP 重解正深度解。补充回归测试。 Co-authored-by: Cursor <cursoragent@cursor.com>	2026-07-21 16:46:05 +08:00
xsl	f870c20f7f	fix(接口3测试页): base64 图片补 data:image/jpeg;base64, 前缀后端返回的 hair_growth_image_base64 是裸 base64（无 data: 前缀），原代码直接赋给 <img>.src，浏览器把 base64 当相对路径请求，拼成 http://host:8187/9j/4AAQ... 发 GET 导致 400 Bad Request。参照接口2 测试页的写法，手动拼接 data:image/jpeg;base64, 前缀。后端代码无需改动（接口2 一直正常）。	2026-07-19 15:14:41 +08:00
xsl	6d9e92f710	chore: torch 升级为 2.11.0+cu128 支持 RTX 5090 (sm_120) 原 torch 2.2.2+cu121 只编到 sm_90，在 RTX 5090 上 GPU 算子报 'no kernel image'，接口2 因 SegFormer/BiSeNet 跑不动 GPU 而失败。 venv 实际已升级到 torch 2.11.0+cu128 / torchvision 0.26.0+cu128，此处同步 requirements.txt 让文档与实际环境一致。 numpy 保持 1.26.4（<2，mediapipe/scikit-image 依赖），cu128 兼容。	2026-07-18 16:30:18 +08:00
xsl	b58cd4c441	优化为4b 模型	2026-07-18 15:30:31 +08:00