docs: 接入页补充接口1/6 的竖线坐标

Co-authored-by: Cursor <cursoragent@cursor.com>
feat(测量): 接口1/6 返回与标注图一致的竖线坐标
2026-11-05 22:29:59 +08:00 · 2026-11-05 22:18:12 +08:00 · 2026-11-04 10:45:16 +08:00 · 2026-11-04 10:20:29 +08:00 · 2026-11-01 15:14:29 +08:00 · 2026-09-20 22:07:48 +08:00
40 changed files with 3641 additions and 71 deletions
@@ -9,6 +9,7 @@ gateway/config.json

 # worker 配置（含鉴权密码，不入 git）
 worker_config.json
+ip_access.json

 # ComfyUI Basic Auth 密码（不入 git）
 password.txt
@@ -2,15 +2,20 @@

 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。
 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。
+worker 对 `/api/*` 校验内网鉴权头 `X-Internal-Token`，`/health` 供网关探测不校验
+（深检：磁盘水位 + 下游依赖 TCP 探测，异常返回 503 degraded，见 _load_health_cfg）。
 """
+import asyncio
 import base64
 import json
 import logging
 import os
+import shutil
+import time
 from contextlib import asynccontextmanager
 from io import BytesIO
 from typing import Any, List, Optional
+from urllib.parse import urlsplit

 import cv2
 import numpy as np
@@ -189,6 +194,10 @@ async def internal_token_auth(request: Request, call_next):
    本地起步），生产务必通过 worker_config.json / 环境变量配置密码。
    """
    path = request.url.path
+    from ip_access import rejection
+    denied = rejection(request)
+    if denied is not None:
+        return denied
    if path == "/" or path.startswith(_AUTH_EXEMPT):
        return await call_next(request)
    if path.startswith("/api/") and ACCEPT_PASSWORDS:
@@ -365,22 +374,22 @@ class ImageJsonBody(BaseModel):
 # ---------------------------------------------------------------------------


-def _run_face_measure_data(image, variant="v1"):
+def _run_face_measure_data(image, variant="v1", landmarks_override=None):
    """接口1测量数值核心：detect → 姿态校验 → 头发/耳朵分割 → measure_face → eye1~eye7。

    返回 (data_dict, result, hair_mask, ear_mask)，或检测/姿态失败时返回 None。
    不含标注图（annotated_image_*），供接口5 容错复用；接口1 调用后再自行生成标注图。
    任何分割/七眼计算异常均内部吞掉（七眼相关字段不出现），不影响主测量结果。

-    variant="v1"（接口1/接口5）：完整四庭七眼 + eye1~eye7（含头部端线段）。
-    variant="v6"（接口6）：去顶庭、不画端线、无 eye1~eye7。
+    variant="v1"（接口1/接口5）：完整四庭七眼 + eye1~eye7（含头部端线段）+ vertical_lines。
+    variant="v6"（接口6）：去顶庭、不画端线、无 eye1/eye7，vertical_lines 仅 6 条内线。
    """
    h, w = image.shape[:2]
    from face_analysis.detector import detector
    from face_analysis.pose import estimate_head_pose, check_frontal_face
    from face_analysis.measure import measure_face

-    landmarks = detector.detect(image)
+    landmarks = landmarks_override if landmarks_override is not None else detector.detect(image)
    if landmarks is None:
        return None
    if not check_frontal_face(landmarks, w, h):
@@ -419,30 +428,29 @@ def _run_face_measure_data(image, variant="v1"):
        # 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。
+    # eye2~eye6（5段）只用内部分点，接口1/6 共用；eye1/eye7 用竖线坐标里的人头端线，仅接口1 有。
+    # vertical_lines 与标注图竖线同一套几何。
    try:
+        from face_analysis.annotation import vertical_line_geometry
        epts = result.eyes["points"]
-        lcx, rcx = epts["left_cheek"][0], epts["right_cheek"][0]
        pc = result.px_per_cm
-        inner_xs = [lcx, epts["left_outer"][0], epts["left_inner"][0],
-                    epts["right_inner"][0], epts["right_outer"][0], rcx]
+        geom = vertical_line_geometry(
+            result, w, h, ear_mask=ear_mask, hair_mask=hair_mask, variant=variant)
+        data["vertical_lines"] = geom
+        inner_xs = [epts["left_cheek"][0], epts["left_outer"][0], epts["left_inner"][0],
+                    epts["right_inner"][0], epts["right_outer"][0], epts["right_cheek"][0]]
        for i in range(5):
            a, b = inner_xs[i], inner_xs[i + 1]
            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。
-            # 竖向范围上界用头顶（hair_top 始终可用）。
-            from face_analysis.annotation import _ear_edges_from_mask
-            top_y = vd["hair_top"][1]
-            head_l, head_r = _ear_edges_from_mask(
-                ear_mask, hair_mask,
-                top_y, vd["chin_tip"][1],
-                lcx, rcx, (lcx + rcx) / 2)
+            by_name = {line["name"]: line["x"] for line in geom["lines"]}
+            head_l = by_name.get("left_head_edge")
+            head_r = by_name.get("right_head_edge")
            data["seven_eyes"]["eye1"] = (
-                None if (head_l is None) else round((lcx - head_l) / pc, 2))
+                None if head_l is None else round((by_name["left_cheek"] - head_l) / pc, 2))
            data["seven_eyes"]["eye7"] = (
-                None if (head_r is None) else round((head_r - rcx) / pc, 2))
+                None if head_r is None else round((head_r - by_name["right_cheek"]) / pc, 2))
    except Exception as seg_e:  # noqa: BLE001
        logger.warning("七眼段宽度计算失败：%s", seg_e)

@@ -892,6 +900,7 @@ async def _face_measure_impl(image_file, image_url, image_base64, variant="v1"):
 - 四庭（顶庭/上庭/中庭/下庭）各段**厘米数值及占比**
 - 七眼（眼宽/脸宽/两眼间距）**厘米数值及占比**
 - 五个关键分界点的**原图像素坐标**（头顶/发际线/眉心/鼻翼下缘/下巴尖）
+- 竖线坐标 `vertical_lines`（与标注图同一套：共用 y 范围 + 从左到右的 x）

 {_image_fields_desc}

@@ -947,6 +956,19 @@ async def _face_measure_impl(image_file, image_url, image_base64, variant="v1"):
                                "nose_bottom": {"x": 540, "y": 1050},
                                "chin_tip": {"x": 540, "y": 1360},
                            },
+                            "vertical_lines": {
+                                "y_top": 98, "y_bottom": 1410,
+                                "lines": [
+                                    {"name": "left_head_edge", "x": 120},
+                                    {"name": "left_cheek", "x": 210},
+                                    {"name": "left_eye_outer", "x": 310},
+                                    {"name": "left_eye_inner", "x": 420},
+                                    {"name": "right_eye_inner", "x": 560},
+                                    {"name": "right_eye_outer", "x": 670},
+                                    {"name": "right_cheek", "x": 780},
+                                    {"name": "right_head_edge", "x": 890},
+                                ],
+                            },
                        },
                    }
                }
@@ -1023,6 +1045,7 @@ async def face_measure_debug(
 - 三庭（上庭/中庭/下庭）各段**厘米数值及占比**（**不含顶庭**）
 - 五眼段宽（eye2~eye6：左脸颊/左眼/两眼间距/右眼/右脸颊）**厘米数值**，另含眼宽/脸宽/两眼间距
 - 四个关键分界点的**原图像素坐标**（发际线/眉心/鼻翼下缘/下巴尖）
+- 竖线坐标 `vertical_lines`（发际线→下巴尖，仅 6 条内线，无人头两端）

 基于接口1 的变体，与接口1 的差异：
 - **去顶庭**：不画头顶横线、不返回顶庭数据；`face_total_height_cm` 为三庭之和
@@ -1080,6 +1103,17 @@ async def face_measure_debug(
                                "nose_bottom": {"x": 540, "y": 1050},
                                "chin_tip": {"x": 540, "y": 1360},
                            },
+                            "vertical_lines": {
+                                "y_top": 430, "y_bottom": 1360,
+                                "lines": [
+                                    {"name": "left_cheek", "x": 210},
+                                    {"name": "left_eye_outer", "x": 310},
+                                    {"name": "left_eye_inner", "x": 420},
+                                    {"name": "right_eye_inner", "x": 560},
+                                    {"name": "right_eye_outer", "x": 670},
+                                    {"name": "right_cheek", "x": 780},
+                                ],
+                            },
                        },
                    }
                }
@@ -1552,7 +1586,7 @@ async def face_features(


 # ---------------------------------------------------------------------------
-# 接口 5：发际线 PNG 生成
+# 接口 5：发际线 PNG 生成（恢复 a3fddbb 历史版本）
 # ---------------------------------------------------------------------------

@app.post(
@@ -1722,6 +1756,285 @@ async def hairline_generate(
        return err(1007, f"处理失败：{ex}")


+# ---------------------------------------------------------------------------
+# 接口 52：中下庭规则、扶正裁切与矩形叠加预览
+# ---------------------------------------------------------------------------
+
+async def _generate_hairline52_data(image, gender, hair_styles, use_mask=True,
+                                        prompt=None, generate_grow_image=False,
+                                        compare_rules=False, normalize_image=True):
+    """Share measurement coordinates between interface 52 and its comparison page."""
+    from fastapi.concurrency import run_in_threadpool
+    from hairline.service import generate_hairline_pngs_52, get_landmarker
+
+    original = image
+    framing, normalized_landmarks, landmarks_override = None, None, None
+    if normalize_image:
+        from hairline.framing import normalize_portrait
+        from mediapipe.framework.formats import landmark_pb2
+        landmarks = await run_in_threadpool(
+            get_landmarker().detect, cv2.cvtColor(image, cv2.COLOR_BGR2RGB), include_iris=True)
+        if landmarks is None:
+            return None, None
+        image, normalized_landmarks, framing = await run_in_threadpool(normalize_portrait, image, landmarks)
+        landmarks_override = landmark_pb2.NormalizedLandmarkList()
+        for point in normalized_landmarks:
+            landmarks_override.landmark.add(x=float(point[0]), y=float(point[1]), z=float(point[2]))
+    fm, reference = None, None
+    try:
+        measured = await run_in_threadpool(_run_face_measure_data, image, landmarks_override=landmarks_override)
+        if measured is not None:
+            fm, result = measured[:2]
+            reference = {key: result.vertical[key]
+                         for key in ("brow_center", "nose_bottom", "chin_tip")}
+            reference["px_per_cm"] = result.px_per_cm
+    except Exception as ex:
+        logger.warning("接口52 当前底图测量失败，使用发际线关键点计算中下庭：%s", ex)
+    res = await run_in_threadpool(
+        generate_hairline_pngs_52, image, gender, hair_styles, use_mask, prompt,
+        generate_grow_image=generate_grow_image, height_reference=reference,
+        compare_rules=compare_rules,
+        precomputed_landmarks=normalized_landmarks[:468] if normalized_landmarks is not None else None)
+    if res is not None:
+        res.update(processed_bgr=image, original_bgr=original, framing=framing)
+    return res, fm
+
+
+@app.post(
+    "/api/v1/hairline/generate-52",
+    summary="接口52 发际线PNG生成",
+    tags=["人脸分析"],
+    description=f"""
+输入用户照片 + 性别 + 多选发型，各发型 middle 中点统一对齐到眉心上方（中庭+下庭）/2。
+返回设计曲线、生发图、目标高度与定位误差；high/low 旧字段保留为 null。
+
+{_image_fields_desc}
+
+图片同时支持 `multipart/form-data` 文件上传（字段名 `image_file`）。
+
+---
+
+**入参**（同接口2：先选性别，再多选发型）：
+- 必填 `gender`（`male`/`female`），决定发型集合（female 7 / male 7）。
+- 必填 `hair_style`（发型序号，逗号分隔如 `1,2,3`），决定返回哪些发际线类型。缺失/越界/非法返回 `1007`。
+  `female`：1=ellipse,2=flower,3=heart,4=straight,5=wave,6=bigflower,7=clasicalflower；
+  `male`：1=ellipse,2=inverse_arc,3=m,4=straight,5=heart,6=Softpetal,7=wave。
+- 可选 `use_mask` / `prompt`：同接口2 的生发控制参数。
+  生发遮罩直接使用新位置的 middle 曲线，与预览保持一致。
+- 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`，仅返回设计叠图与中心点。
+- 新增可选 `normalize_image`（默认 `false`）：显式传 `true` 时扶正并裁切，长图采用3:4。
+  不传时保留旧调用的原图尺寸与坐标。测试页显式开启此参数，保留实际像素、不放大图片。
+
+**返回说明**：
+
+- `hairline_images`：**选中发型**的列表，数量 = 所选发型数，`order` = 发型序号。每项含：
+  - `image_middle_base64`：新高度规则的透明发际线 PNG；`image_high_base64` / `image_low_base64` 固定 null。
+  - `height_placement`：旧/新中心坐标、移动距离和目标误差（像素）。
+  - `grown_image_base64`：该发型的**生发图**（生发失败时为 `null`）。
+  - `hairline_type`：发际线类型 key。
+  本接口仅部署在GPU worker，图片始终返回 `*_base64`；空图片保留字段为 null。
+- `best_hairline_center_point`：**首个选中发型**的 **middle 档**发际线曲线**面部中间点**坐标，
+  以**当前底图像素**为基准（左上角为原点，x 向右，y 向下）。
+- `high_hairline_center_point` / `low_hairline_center_point`：旧字段，固定 null。
+- `height_rule`：中庭、下庭、目标上庭高度、目标中心与测量来源；`face_measure` 为当前底图测量。
+- `input_image_base64`：处理后底图，须用它叠加发际线；`original_image_base64`：处理前图片。
+- `coordinate_space`：默认 `original_image`，开启预处理后为 `processed_image`。
+- `framing`：是否标准裁切、旋转角、前后尺寸及双向坐标变换；关闭预处理时为 null。
+""",
+    responses={
+        200: {
+            "description": "成功",
+            "content": {
+                "application/json": {
+                    "example": {
+                        "code": 0,
+                        "message": "success",
+                        "request_id": "mock-request-id",
+                        "data": {
+                            "hairline_images": [
+                                {"hairline_type": "ellipse", "image_middle_base64": "iVBORw0KGgo...", "image_high_base64": None, "image_low_base64": None, "grown_image_base64": "iVBORw0KGgo...", "order": 1},
+                                {"hairline_type": "flower", "image_middle_base64": "iVBORw0KGgo...", "image_high_base64": None, "image_low_base64": None, "grown_image_base64": None, "order": 2},
+                            ],
+                            "best_hairline_center_point": {"x": 540, "y": 430},
+                            "high_hairline_center_point": None,
+                            "low_hairline_center_point": None,
+                            "face_measure": {
+                                "face_total_height_cm": 13.76,
+                                "four_courts": {
+                                    "top_court_cm": 3.44, "upper_court_cm": 3.44,
+                                    "middle_court_cm": 3.44, "lower_court_cm": 3.44,
+                                    "ratios": {"top_court": 0.25, "upper_court": 0.25,
+                                               "middle_court": 0.25, "lower_court": 0.25},
+                                },
+                                "seven_eyes": {
+                                    "eye_width_cm": 3.44, "face_width_cm": 24.08,
+                                    "inter_eye_distance_cm": 3.44,
+                                    "ratios": {"eye_width": 0.143, "inter_eye_distance": 0.143},
+                                    "eye1": 3.44, "eye2": 3.44, "eye3": 3.44, "eye4": 3.44,
+                                    "eye5": 3.44, "eye6": 3.44, "eye7": 3.44,
+                                },
+                                "landmarks": {
+                                    "hair_top": {"x": 540, "y": 120},
+                                    "hairline": {"x": 540, "y": 430},
+                                    "brow_center": {"x": 540, "y": 740},
+                                    "nose_bottom": {"x": 540, "y": 1050},
+                                    "chin_tip": {"x": 540, "y": 1360},
+                                },
+                                "hairline_source": "segmentation",
+                                "head_pose": {"yaw": -1.39, "pitch": 2.49, "roll": -0.06},
+                            },
+                        },
+                    }
+                }
+            },
+        },
+        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 hairline_generate_52(
+    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:1-7 male:1-7"),
+    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="填充遮罩区域的头发，人物增加一点磨皮", 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。false 时仅返回新规则发际线叠图与中心点"),
+    normalize_image: bool = Form(default=False, description="新增可选开关，默认false保留原图尺寸和坐标；true先扶正再裁切，长图采用3:4。测试页显式开启"),
+):
+    if gender not in ("male", "female"):
+        return err(1004, "gender 必填且只能为 male / female")
+
+    # hair_style 必填（同接口2）：解析逗号分隔，缺失/越界/非法 → 1007
+    max_styles = {"female": 7, "male": 7}[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}")
+
+    raw, e = await resolve_image_bytes(image_file, image_url, image_base64)
+    if e is not None:
+        return e
+
+    from hairline.framing import decode_portrait
+    image = decode_portrait(raw)
+    if image is None:
+        return err(1008, "图片格式不支持（仅 JPG / PNG）")
+
+    try:
+        res, fm = await _generate_hairline52_data(
+            image, gender, hair_styles, use_mask, prompt,
+            generate_grow_image=generate_grow_image, normalize_image=normalize_image)
+        if res is None:
+            return err(1001, "无法识别人像")
+
+        hairline_images = []
+        for it in res["images"]:
+            ov = it["overlays"]
+            hairline_images.append({
+                "hairline_type": it["hairline_type"],
+                "image_middle_base64": _rgba_png_b64(ov["middle"]),
+                "image_high_base64": None,
+                "image_low_base64": None,
+                "grown_image_base64": (_png_to_jpg_b64(it["grown_png"]) # 生发图 JPG（失败为 null）
+                                       if it.get("grown_png") else None),
+                "order": it["order"],
+                "height_placement": it["height_placement"],
+            })
+        c = res.get("best_centers") or {}
+        def _pt(p):
+            return ({"x": p[0], "y": p[1]} if p else None)
+        data = {
+            "hairline_images": hairline_images,
+            "best_hairline_center_point": _pt(c.get("middle")),
+            "high_hairline_center_point": _pt(c.get("high")),
+            "low_hairline_center_point": _pt(c.get("low")),
+            "height_rule": res["height_rule"],
+            "guide_landmarks": res["guide_landmarks"],
+            "face_measure": fm,
+            "framing": res["framing"],
+            "input_image_base64": _jpg_b64(res["processed_bgr"]),
+            "original_image_base64": _jpg_b64(res["original_bgr"]),
+            "coordinate_space": "processed_image" if normalize_image else "original_image",
+            "width": res["processed_bgr"].shape[1], "height": res["processed_bgr"].shape[0],
+        }
+        return ok(data)
+    except Exception as ex:  # noqa: BLE001
+        logger.exception("接口52 处理异常")
+        return err(1007, f"处理失败：{ex}")
+
+
+@app.get("/api/v1/debug/hairline-height-cases", include_in_schema=False)
+async def hairline_height_cases():
+    path = os.path.join(os.path.dirname(__file__), "tests", "fixtures", "hairline_height_cases.json")
+    with open(path, encoding="utf-8") as f:
+        return ok(json.load(f))
+
+
+@app.get("/api/v1/debug/hairline-framing-reference", include_in_schema=False)
+async def hairline_framing_reference():
+    from hairline.framing import load_profile
+    return ok(load_profile())
+
+
+@app.post("/api/v1/debug/hairline-height-compare", include_in_schema=False)
+async def hairline_height_compare(
+    image_file: Optional[UploadFile] = File(default=None),
+    image_url: Optional[str] = Form(default=None),
+    image_base64: Optional[str] = Form(default=None),
+    gender: str = Form(default="female"),
+    hair_style: str = Form(default="1,2,3,4,5,6,7"),
+    generate_grow_image: bool = Form(default=False),
+    normalize_image: bool = Form(default=True),
+):
+    if gender not in ("female", "male"):
+        return err(1004, "gender 必须为 male / female")
+    styles = _parse_hair_styles(hair_style, 7)
+    if styles is None:
+        return err(1007, "hair_style 必须为 1..7 的整数，逗号分隔")
+    raw, error = await resolve_image_bytes(image_file, image_url, image_base64)
+    if error is not None:
+        return error
+    from hairline.framing import decode_portrait
+    photo = decode_portrait(raw)
+    if photo is None:
+        return err(1008, "图片格式不支持")
+    try:
+        res, fm = await _generate_hairline52_data(
+            photo, gender, styles, prompt="填充遮罩区域的头发，人物增加一点磨皮",
+            generate_grow_image=generate_grow_image, compare_rules=True,
+            normalize_image=normalize_image)
+        if res is None:
+            return err(1001, "无法识别人像")
+        h, w = res["processed_bgr"].shape[:2]
+        return ok({
+            "gender": gender, "width": w, "height": h,
+            "input_base64": _jpg_b64(res["processed_bgr"]),
+            "original_base64": _jpg_b64(res["original_bgr"]),
+            "framing": res["framing"],
+            "coordinate_space": "processed_image" if normalize_image else "original_image",
+            "height_rule": res["height_rule"], "face_measure": fm,
+            "guide_landmarks": res["guide_landmarks"],
+            "results": [{
+                "hairline_type": item["hairline_type"], "order": item["order"],
+                "old_overlay_base64": _rgba_png_b64(item["legacy_overlay"]),
+                "new_overlay_base64": _rgba_png_b64(item["overlays"]["middle"]),
+                "height_placement": item["height_placement"],
+                "grown_image_base64": _png_to_jpg_b64(item["grown_png"]) if item["grown_png"] else None,
+            } for item in res["images"]],
+        })
+    except Exception as ex:
+        logger.exception("接口52 高度规则对比失败")
+        return err(1007, str(ex))
+
+
 # ---------------------------------------------------------------------------
 # 接口 9：头发遮罩生成
 # ---------------------------------------------------------------------------
@@ -2328,12 +2641,127 @@ async def download_hairline_log(rid: Optional[str] = None, tail: int = 500):
 # 健康检查
 # ---------------------------------------------------------------------------

+def _load_health_cfg() -> dict:
+    """深健康检查配置：环境变量优先，回退 worker_config.json 的 "health" 段。
+
+    min_free_gb=0 关闭磁盘检查；check_deps=false 关闭依赖探测（均退化为旧行为，
+    即只看模型就绪标志）。
+    """
+    cfg_path = os.path.join(os.path.dirname(__file__), "worker_config.json")
+    file_cfg: dict = {}
+    if os.path.isfile(cfg_path):
+        try:
+            with open(cfg_path, encoding="utf-8") as f:
+                file_cfg = json.load(f).get("health") or {}
+        except Exception as e:  # noqa: BLE001
+            logger.warning("读取 worker_config.json health 段失败：%s", e)
+
+    def _get(env_key, file_key, default):
+        raw = os.getenv(env_key)
+        return raw if raw not in (None, "") else file_cfg.get(file_key, default)
+
+    def _as_float(value, default):
+        try:
+            return float(value)
+        except (TypeError, ValueError):
+            return default
+
+    return {
+        "min_free_gb": _as_float(_get("HEALTH_MIN_FREE_GB", "min_free_gb", 5.0), 5.0),
+        "check_deps": str(_get("HEALTH_CHECK_DEPS", "check_deps", True)).strip().lower()
+                      not in ("0", "false", "no", "off"),
+        "dep_comfyui": str(_get("HEALTH_DEP_COMFYUI", "dep_comfyui", "127.0.0.1:8188")),
+        "deps_cache_seconds": max(0.0, _as_float(
+            _get("HEALTH_DEPS_CACHE_SECONDS", "deps_cache_seconds", 5.0), 5.0)),
+    }
+
+
+_HEALTH_CFG = _load_health_cfg()
+
+
+def _disk_free_gb() -> float:
+    """worker 所在文件系统的剩余空间（GB）。"""
+    st = shutil.disk_usage(os.path.dirname(os.path.abspath(__file__)))
+    return st.free / (1 << 30)
+
+
+def _parse_host_port(addr: str, default_port: int):
+    """解析 "host:port"（仅支持 IPv4/主机名形式）。"""
+    host, _, port = str(addr).partition(":")
+    try:
+        return host or "127.0.0.1", int(port) if port else default_port
+    except ValueError:
+        return host or "127.0.0.1", default_port
+
+
+def _change_hair_addr():
+    """从 _CHANGE_HAIR_BASE 解析 (host, port)，失败返回 None。"""
+    try:
+        u = urlsplit(_CHANGE_HAIR_BASE)
+        if u.hostname:
+            return u.hostname, u.port or 80
+    except ValueError:
+        pass
+    return None
+
+
+async def _tcp_probe(host: str, port: int, timeout: float = 0.8) -> bool:
+    """TCP 连通性探测：只判下游进程死活，不过 HTTP 认证、不受业务繁忙影响。"""
+    try:
+        _, writer = await asyncio.wait_for(asyncio.open_connection(host, port), timeout)
+        writer.close()
+        return True
+    except Exception:  # noqa: BLE001
+        return False
+
+
+_DEP_CACHE = {"ts": -1.0, "detail": None}
+
+
+async def _check_deps():
+    """并发探测下游依赖（ComfyUI、change_hair），结果缓存 deps_cache_seconds 秒。"""
+    now = time.monotonic()
+    detail = _DEP_CACHE["detail"]
+    if detail is None or now - _DEP_CACHE["ts"] >= _HEALTH_CFG["deps_cache_seconds"]:
+        probes = {"comfyui": _tcp_probe(*_parse_host_port(_HEALTH_CFG["dep_comfyui"], 8188))}
+        ch = _change_hair_addr()
+        if ch:
+            probes["change_hair"] = _tcp_probe(*ch)
+        names = list(probes)
+        results = await asyncio.gather(*probes.values())
+        detail = dict(zip(names, results))
+        _DEP_CACHE["ts"], _DEP_CACHE["detail"] = now, detail
+    return all(detail.values()), detail
+
+
@app.get("/health", include_in_schema=False)
 async def health():
-    # 模型就绪后才返回 200，供网关探测；未就绪返回 503。
+    """网关探测入口。
+
+    模型未就绪 → 503 starting；磁盘剩余低于阈值 / 下游依赖 TCP 不通 → 503 degraded；
+    正常 → 200（附 disk_free_gb 与 deps 明细，供人工排查）。深检自身异常时按就绪返回
+    （健康检查不能自己成为故障点）。
+    """
    if not _STATE["ready"]:
        return JSONResponse(status_code=503, content={"status": "starting"})
-    return {"status": "ok"}
+    try:
+        free_gb = _disk_free_gb()
+        body = {"status": "ok", "disk_free_gb": round(free_gb, 1)}
+        min_free = _HEALTH_CFG["min_free_gb"]
+        if min_free > 0 and free_gb < min_free:
+            body.update(status="degraded", reason=f"disk_free_below_{min_free}g")
+            return JSONResponse(status_code=503, content=body)
+        if _HEALTH_CFG["check_deps"]:
+            deps_ok, detail = await _check_deps()
+            body["deps"] = detail
+            if not deps_ok:
+                down = ",".join(k for k, v in detail.items() if not v)
+                body.update(status="degraded", reason=f"dep_down:{down}")
+                return JSONResponse(status_code=503, content=body)
+        return body
+    except Exception:  # noqa: BLE001
+        logger.exception("深健康检查异常，按就绪返回")
+        return {"status": "ok"}


@app.get("/", include_in_schema=False)
@@ -13,21 +13,22 @@
 ```
 客户端 ──HTTPS──> 外网网关(gateway/)  ──HTTP(X-Internal-Token)──> worker(GPU 机, app.py)
                     │ 薄代理 + 落盘改URL                          │ 跑算法(本地模型/ComfyUI/change_hair)
-                     └ 接口4(豆包)/99(特殊点) 本机实现(不转发)      └ 接口1/2/3/5/6/7/9/10/11/12 + 工具路由
+                     └ 接口4(豆包)/99(特殊点) 本机实现(不转发)      └ 接口1/2/3/5/52/6/7/9/10/11/12 + 工具路由
 ```

 - **worker**（`app.py` + `face_analysis/` + `hairline/`）：跑真正的算法，除 ComfyUI/change_hair 走本机
  回环外无外网依赖。对 `/api/*` 校验 `X-Internal-Token`（密码在 `worker_config.json` 的 `accept_passwords`）；
  `/health` 模型就绪才返回 200。监听 **8187**（systemd `hair-worker.service`，也可 `./run_worker.sh` 手动起）。
  另有 worker 本机的工具路由：`/api/swapHair/v1`、`/hairColor/v2`（CORS 代理转发 change_hair:8801）、
-  `/api/v1/redraw`（直接 ComfyUI 重绘）、`/api/v1/debug/*`（免鉴权调试）。
+  `/api/v1/redraw`（直接 ComfyUI 重绘）、`/api/v1/debug/*`（免Token调试，仍受IP限制）。
+  所有动态路径先校验直连来源IP：允许本机与网关`49.232.51.171`，其余返回HTTP403/code1010；`/static/`除外。配置见[`接口IP白名单.md`](接口IP白名单.md)，网关侧不增加IP限制。
 - **网关**（`gateway/`）：薄反向代理，健康检查/派发/鉴权/把 worker 的 `*_base64` 落盘改成 `*_url`。
  网关本机实现的接口（不转发 worker）：**接口4**（调外网豆包）与**接口99**（纯 CPU 特殊点检测，
  直接 import 同仓库 `face_analysis/special_points`，`_if99_lock` 单路排队）。
 - **图片三选一**：所有接口图片入参 `image_file`/`image_url`/`image_base64` 严格三选一（接口3 是 `marked_image_*`）。
- **响应信封**：`{code, message, request_id, data}`；业务错误用 `code`（HTTP 一律 200）。
+- **响应信封**：`{code, message, request_id, data}`；业务错误用 `code`（HTTP 200）；IP拒绝为HTTP403/code1010，Token错误为HTTP401/code1009，依赖健康异常为HTTP503。

-### base64 → URL 映射（网关落盘改写，递归进数组、可空保留 null）
+### base64 → URL 映射（网关落盘改写，递归进数组）

 | 接口 | worker 字段（内部） | 对外字段 |
 |------|--------------------|----------|
@@ -36,6 +37,7 @@
 | 2 | `results[].image_base64` / `results[].grown_image_base64`(可空) | `results[].image_url` / `results[].grown_image_url` |
 | 3 | `hair_growth_image_base64`(可空) | `hair_growth_image_url` |
 | 5 | `hairline_images[].image_{middle,high,low}_base64` / `grown_image_base64`(可空) | `hairline_images[].image_{middle,high,low}_url` / `grown_image_url` |
+| 52 | `hairline_images[].image_{middle,high,low}_base64` / `grown_image_base64`、`input_image_base64` / `original_image_base64` | 仅GPU worker，本次不新增网关路由；直接返回`*_base64` |
 | 9 | `steps_common.*_base64`、`bisenet|segformer.{hair,closed_region,final_overlay,mask}_mask_base64` 等 | 同名 `*_url` |
 | 10 | `steps_common.*_base64`、`bisenet|segformer.{base_mask,contour,band_overlay,mask}_base64` 等 | 同名 `*_url` |
 | 11 | `steps.{input,baseline,hair_mask,mask_overlay,mask,swap_raw,alpha,final}_base64` | `steps.*_url` |
@@ -43,20 +45,21 @@
 | 4 | （网关本机产出，无图片字段，`features` 为 JSON 字符串） | — |

 > 实现建议：递归遍历 data，凡 key 以 `_base64` 结尾就落盘改 `_url`，自动覆盖嵌套/新增字段。
+> 兼容约定：网关保持原样，有图片时落盘为`*_url`，空图片保留`*_base64: null`。接口52仅在worker返回base64，与网关的图片改名无关。
 > **图片格式**：接口1 标注图含透明用 **PNG**；接口2/3/5 是不透明照片用 **JPG**(小很多，~9×)。
 > 网关落盘按内容嗅探扩展名（PNG 头→`.png`，否则 `.jpg`）。

 ### 错误码

 `1001` 无法识别人像 ｜ `1002` 分辨率过低 ｜ `1003` 非正面 ｜ `1004` gender 必填/非法（接口2/5/7）｜
-`1006` >1MB ｜ `1007` 图片参数错误（0或多个）/未预期异常 ｜ `1008` 格式不支持。
+`1006` >1MB ｜ `1007` 图片参数错误（0或多个）/未预期异常 ｜ `1008` 格式不支持 ｜ `1009` Token鉴权失败 ｜ `1010` worker来源IP不允许或白名单配置无效。

 ---

 ## 2. 各接口实现简述

 ### 接口1 四庭七眼测量 `/api/v1/face/measure`（worker）
- **做什么**：正面照 → 四庭(顶/上/中/下庭) + 七眼(眼宽/脸宽/间距) 的 cm 与占比、5 个关键点坐标、一张透明底标注 PNG。
+- **做什么**：正面照 → 四庭(顶/上/中/下庭) + 七眼(眼宽/脸宽/间距) 的 cm 与占比、5 个关键点坐标、竖线坐标 `vertical_lines`、一张透明底标注 PNG。
 - **怎么实现**（`face_analysis/`）：MediaPipe Face Mesh 468+虹膜点 → solvePnP 姿态校验(非正面 1003) →
  虹膜直径法定标(px→cm) → **眉心以下实测**；**眉心以上**用 BiSeNet 头发分割取真实发际线/头顶(方案B)，
  失败回退比例推算(方案A，`hairline_source` 透出)。标注图 numpy 向量化渐变线 + 思源黑体。返回 `annotated_image_base64`。
@@ -65,7 +68,7 @@
 ### 接口6 四庭七眼测量 v2 `/api/v1/face/measure-v2`（worker）—— 接口1 的去顶庭变体
 - **做什么**：基于接口 1，**去顶庭**：不画头顶横线、不返回顶庭数据（`four_courts` 仅上/中/下庭，`landmarks` 无 `hair_top`，`face_total_height_cm` 为三庭之和）。
 - **与接口1 的标注差异**（`create_annotated_image(variant="v6")`）：①竖线范围改为发际线→下巴尖；②不画人头最左/最右端线（仅七眼 6 点 5 段，接口1 为 8 线 7 段）；③左侧只标上/中/下庭。箭头/虚线/字体等与接口1 一致。
- **怎么实现**：与接口 1 共用 `_face_measure_impl(variant="v6")`；v6 时标注走变体分支、数据由 app.py 边界删顶庭字段并重算三庭比例。
+- **怎么实现**：与接口 1 共用 `_face_measure_impl(variant="v6")`；v6 时标注走变体分支、数据由 app.py 边界删顶庭字段并重算三庭比例。`vertical_lines` 与标注图共用 `vertical_line_geometry`：接口 6 的 y 范围是发际线到下巴尖，`lines` 只有 6 条内线。
 - **网关改动**：新增路由 `POST /api/v1/face/measure-v2`，转发到 worker 同路径；base64→URL 改写无需改动。

 ### 接口2 C端生发 `/api/v1/hair/grow`（worker）—— 预览 + 生发图
@@ -74,6 +77,7 @@
  按 `face_ext.obj` 的 UV 把发际线贴图渲染到额头(预览)。原生生发：黑贴图渲染遮罩 → 调本机 **ComfyUI 8188** 的
  `add_hair.json`(Flux-2) 出图。**关键坑**：obj 是重排序，需 `INDEX_MAP_468` 把 MP 序→OBJ 序。
  返回 `results[].image_base64` + `grown_image_base64`。
+- 模板对齐：左右刘海边界高低不同时，按 MediaPipe 左右对应额头点协调沿人脸向上方向的延伸距离，避免把 M 型等设计模板拉歪；保留人脸姿态，之后仍做头部轮廓钳制。预览白线与生发黑线/遮罩共用修正后的 502 点几何，不改变真实发际线测量算法。
 - `hair_style` 映射：female 1=ellipse 2=flower 3=heart 4=straight 5=wave 6=bigflower 7=clasicalflower；male 1=ellipse 2=inverse_arc 3=m 4=straight 5=heart 6=Softpetal 7=wave。female 1..5 走「换发型(change_hair)」+Flux-2 重绘管线（`generate_grow_results_swap`，复用接口12 final 的 `_V2_FINAL_DEFAULTS`，长边默认压到 `REDRAW_MAX_SIDE`=1024 再放大回原尺寸）；female 6/7 与 male 全部走原生发(ComfyUI add_hair)管线。可选参数 `flux_model`（切换 Flux unet，.gguf 自动配 UnetLoaderGGUF + 配套 qwen CLIP）、`redraw_max_side`（0=不缩图）。

 ### 接口7 C端生发 v2 `/api/v1/hair/grow-v2`（worker）—— 接口2同款，add_hair2 工作流
@@ -98,6 +102,15 @@
 - **做什么**：入参同接口2（`gender` + 多选 `hair_style` 必填）。对每个选中发型 → `middle`/`high`/`low` 三档发际线叠图 + 生发图 + 首个选中发型的面部中间点坐标。
 - **怎么实现**：复用接口2 的 502 点渲染管线，三档分别用 `hairline_texture[/_high|/_low]` 同名贴图渲染叠图；生发同接口2（ComfyUI inpaint），**黑模板固定取 `hairline_texture_black/`（middle）**，每发型 1 张生发图。`best_hairline_center_point`=眉心 x × 首个选中发型 middle 档发际线 y。

+### 接口52 发际线PNG生成 `/api/v1/hairline/generate-52`（worker）
+- **预处理**：`hairline/framing.py`先扶正9→152，再按高(3÷0.9)L、宽2.5L（3:4）做原像素裁切，人物相对原3L规则缩小约10%。L为9→152距离；眉心位置沿用已扶正参考图比例。长图的标准窗口不足时，求原图内最大的3:4窗口，通过可行区域约束保留全部人脸关键点，并尽量贴近目标眉心位置；无法保留人脸时回退仅扶正。其余图片空间不足时仍仅扶正、去除旋转空角。参考图与参数由`scripts/calibrate_hairline_framing.py`生成。公开接口为兼容旧调用默认关闭预处理，`normalize_image=true`开启；规则页显式开启，对比调试页默认开启。返回当前底图、原图及双向坐标矩阵。
+- **目标高度**：`上庭=(中庭+下庭)/2`，用当前底图测量的未取整像素坐标（眉心9、鼻翼下缘94、下巴152）算目标中心。测量失败时用发际线管线同编号点计算并标注来源。
+- **模板定位**：原有502点/贴图生成形状，`hairline/height_rule.py` 将每种 middle 曲线沿面部向上方向平移、校准中心 Y，保留曲线形状与倾斜。此规则仅接口52使用，接口2保持原逻辑。
+- **生发**：移动后的预览曲线直接生成黑线和重绘遮罩，再送ComfyUI，避免预览与黑模板位置不一致。
+- **裁切叠加预览**：测试页新增“裁切发际线叠加”。`static/hairline_reference_box.js`共享同一个162/389/9/152参考框几何，按原图像素裁切middle透明图层，再叠加到处理后人像；四栏顺序为原图、middle、裁切发际线叠加、生发图。
+- **兼容**：high/low不再生成，图片与中心点字段保留为null。接口52始终返回`*_base64`，空图片保留字段并返回null。本次不修改网关路由、落盘、字段改名或IP策略。`face_measure`为当前底图测量；新增`height_rule`与各项`height_placement`说明设计目标与实际误差。
+- **测试**：规则页`static/test_interface52.html`，对比页`static/test_interface52_compare.html`；运行`scripts/compare_hairline_height.py`默认保存全部30张×男女各7型，共420组对比。每人一行，顺序为未裁切原图、单张旧规则图、全部新规则样式，支持位置对照、额头特写、同图叠线和上传新图。新图可叠加162/389两侧、底边经过9、高度为9→152距离70%的参考框。原图与叠图按需加载。
+
 ### 接口9 头发遮罩 `/api/v1/head/mask`（worker）
 - **做什么**：人像照 → 头发遮罩 + 全流程分步可视化。
 - **怎么实现**（`face_analysis/head_mask.py`）：MediaPipe 底部分割线（21/68/104/69/108/151/337/299/333/298/251 连线，21/251 水平延伸）→ 线以上为上半区 → BiSeNet 与 SegFormer 双路头发分割、逐列向下填充到分割线得闭合区域（头发+额头）→ 外缘朝 151 内缩 `erode_cm`(默认1.2cm，虹膜标定) → 最终遮罩。返回 `steps_common` + `bisenet`/`segformer` 两组各四张图。
@@ -125,6 +138,7 @@
 ## 3. 部署 / 环境要点

 **worker（GPU 机，本机 = gpu_worker 角色）**
+- 启动使用`./run_worker.sh`或`hair-worker.service`，均显式传`--no-proxy-headers`以保留直连来源IP。部署时复制`ip_access.example.json`为`ip_access.json`并核对网关IP；运行配置不入Git，改动即时生效。旧启动方式也应补上此参数，详见[`接口IP白名单.md`](接口IP白名单.md)。
 - Python **3.10.12**（本机 venv 在 `./venv`）；`numpy<2`(1.26.4)，`scikit-image==0.24.0`
  （**别升 0.25+，会顶 numpy≥2 顶崩 mediapipe**）。
 - 本机 **RTX 4090**：venv 实装 `torch 2.7.1+cu128`，CUDA 可用（BiSeNet/SegFormer GPU 推理）。
@@ -0,0 +1,120 @@
+# 接口52上传图片裁切与扶正规划
+
+状态：2026-11-04 在原3L规则上将人物显示缩小约10%，裁切高度改为(3÷0.9)L≈3.3333L、宽度2.5L，已接入接口52与对比页。参考图先扶正后重新标定，以下旧预检数值仅为历史记录，当前标准以 `hairline/reference/framing_profile.json` 为准。
+
+## 接口拆分
+
+- 接口5已按11月2日前最新提交`a3fddbb`恢复：`POST /api/v1/hairline/generate`，返回middle/high/low三档，测试页`/static/test_interface5.html`。
+- 当前新规则完整迁至接口52：`POST /api/v1/hairline/generate-52`，测试页`/static/test_interface52.html`，30案例对比`/static/test_interface52_compare.html`。接口52保留迁移前所有参数和返回字段，页面显式开启`normalize_image=true`。
+- 原`test_interface5_compare.html`书签跳转至52号对比页；5号与52号的worker处理路由和生成函数独立；52号仅部署在GPU worker，本次不新增网关转发。
+
+## 已确定的要求
+
+- 仅用于接口52与当前旧/新规则对比页。
+- 先扶正：变换后的9号点与152号点保持同一X坐标，9在上、152在下。
+- 原图空间够时按目标人脸比例和位置裁切；标准窗口不足时，偏长图片在保留人脸的前提下裁为3:4，其余仍仅扶正。3:4无法保留完整人脸时不强裁。
+- 输出保留实际像素尺寸，不为了凑参考图尺寸而放大小图。
+- 当前展示：全部30个案例，每个案例一行，未裁切原图放最前，旧规则图仅一张，新规则男女各7型横排，人物图片完整显示。选择新样式时旧图同步对应形状；失败案例仍保留原图并显示原因。
+
+## 已扶正标准（当前生效）
+
+- 标准图：`static/framing_reference.png`，786×1048，3:4。
+- 两点X均为385.18783691；9号Y为360.65437127，152号Y为725.82510480。
+- 相对位置：9号点 x=49.00608612%，y=34.41358504%。
+- 当前裁切高/L=3÷0.9≈3.3333、宽/L=2.5（3:4），L为9→152长度。参考图自身测得的宽/L=2.15241783589、高/L=2.86989044785作为原始标定记录保留；眉心在画面中的位置比例继续使用参考图数值。
+- 校准旋转约0.25962484°，变换后横向误差约5.7e-14像素。参数取自扶正并去空角后的图，不再使用未扶正原图的比例。
+- 重标定脚本：`./venv/bin/python scripts/calibrate_hairline_framing.py <参考图路径>`。
+- 使用同一套原始检测点作刚体变换与后续测量，避免二次检测抖动破坏轴线竖直。
+- 公开接口新增`normalize_image`默认false，旧请求保持原图尺寸和坐标；显式true开启预处理。规则页主动传true，对比调试页默认开启。响应明确返回`coordinate_space`、当前底图和坐标变换。
+
+## 原始参考构图（旧测量）
+
+[用户指定参考图](https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/misc/20261103_user.jpg)
+
+测量采用当前发际线管线的MediaPipe FaceLandmarker：
+
+| 指标 | 测量值 |
+|---|---|
+| 画面 | 792×1056，3:4 |
+| 9号点 | (388.93, 364.62) |
+| 152号点 | (387.27, 729.79) |
+| 9→152长度L | 365.17像素 |
+| 面部轴线距竖直方向 | 约0.26° |
+| 9号点画面相对位置 | x约49.1%，y约34.5% |
+| 152号点画面相对高度 | y约69.1% |
+| 画面宽度/L | 约2.16885 |
+| 画面高度/L | 约2.89180 |
+
+参考图接近竖直。以9号点为定位锚点，用9→152长度作为缩放无关的构图标尺。这里的L仅为眉心到下巴的距离，不是头顶到下巴的全头高度。
+
+## 实施流程
+
+1. 读取上传原图、处理EXIF方向，再检测人脸关键点。
+2. 用9→152方向计算二维旋转，使转换后两点竖直。不改变脸部比例。
+3. 按本人的L算目标裁切尺寸：高(3÷0.9)L、宽2.5L；令9号点位于裁切画面约(49.0061%,34.4136%)处。高度取最接近的4的倍数，宽=高×3/4，不改变原图像素尺度。
+4. 将裁切窗口四角反算回原图，四角均在有效图片区域内才执行标准构图。采用一次仿射变换合并旋转与裁切，降低重复插值。
+5. 标准窗口空间不足时，偏长图片采用原图内最大的3:4有效窗口，约束所有人脸关键点在框内，并尽量靠近目标眉心位置。若无法保留完整人脸则仅扶正并给出原因。其余图片仍仅扶正，最小去除旋转空角。
+6. 后续测量、分割、旧/新发际线和70%参考框使用同一个处理后底图与坐标系。保留原图及原图/处理后坐标转换关系，防止叠图错位。
+7. 页面展示原图、处理后图、旋转角、是否标准裁切、跳过原因与前后尺寸。旧/新规则对比使用同一处理后底图，避免构图变化混入高度规则差异。
+
+## 当前：长图与矩形发际线预览
+
+- 长图即使放不下标准人脸占比窗口，也尝试在原图有效区域裁出最大的3:4窗口；通过可行区域约束保留468个人脸关键点，并尽量贴近参考眉心位置。无法完整保留人脸时回退仅扶正，响应说明原因。
+- 案例7从1080×1920裁为1062×1416，保持9→152竖直。长图7、20、21、22、50、51的男女各7型已更新，84组均成功；对比页30张、378组结果统一使用当前规则。
+- 规则页展示原图、middle、裁切发际线叠加、生发图。新增预览使用共享矩形几何裁切透明发际线，矩形外清空透明，底图仍是完整处理后人像。
+
+## 人物缩小10%的验证（长图规则加入前）
+
+- 裁切高=(3÷0.9)L、宽=2.5L，使同样展示尺寸下的人物线性尺寸约为原3L规则的90%。
+- 案例1由828×1104改为918×1224，人脸相对大小约90.20%，差异来自原像素窗口取整；底图、发际线和实际生发图均为918×1224。
+- 对比页保留全部30张。重新计算了1、3、42、43、44、45；其余48组结果验证新窗口仍超出有效范围，且只扶正后的矩阵和尺寸完全一致，复用原定位图并更新裁切参数。
+- 当前生成378组对比，43、44、45的男女组仍显示原图及定位失败原因。布局、眉心比例和9→152扶正方向保持一致。
+
+## 历史：3L规则的五个案例验证
+
+2026-11-04 按3L规则重跑男女各7型，共70组，失败0组。保留原始像素，不缩放人脸；实际尺寸如下。
+
+| 案例 | 原尺寸 | 处理后尺寸 | 实际处理 |
+|---|---|---|---|
+| 1 | 1088×1406 | 828×1104 | 按3L规则裁切 |
+| 2 | 1088×1406 | 1042×1346 | 空间不足，仅扶正并去除旋转空角 |
+| 3 | 1088×1406 | 810×1080 | 按3L规则裁切 |
+| 4 | 1088×1406 | 1042×1346 | 空间不足，仅扶正并去除旋转空角 |
+| 5 | 1088×1367 | 1071×1346 | 空间不足，仅扶正并去除旋转空角 |
+
+## 历史几何预检（未扶正参考图）
+
+以下为修改前的预检记录，当前生效尺寸见上表。
+
+| 案例 | 原尺寸 | 所需裁切尺寸（约） | 预计处理 |
+|---|---|---|---|
+| 1 | 1088×1406 | 797×1063 | 可按参考构图裁切并扶正 |
+| 2 | 1088×1406 | 1442×1923 | 空间不足，仅扶正 |
+| 3 | 1088×1406 | 782×1042 | 可按参考构图裁切并扶正 |
+| 4 | 1088×1406 | 1465×1953 | 空间不足，仅扶正 |
+| 5 | 1088×1367 | 1463×1950 | 空间不足，仅扶正 |
+
+## 历史：3L规则的全部30张验证
+
+- 按3L规则运行全部30张、男女各7型，共60次请求；55组请求成功，生成385组旧/新规则对比。27张男女均成功，案例43额外生成男性7型。
+- 案例43女性：模板在面部中轴处没有有效曲线；案例44、45男女：扶正后无法保留9/152参考点。三个案例均在对应行显示完整原图和未生成组合的原因，不隐藏失败案例。
+- 页面顺序为未裁切原图、旧规则、全部新规则；案例筛选包括失败图片，重跑按钮运行完整30张。
+- 页面筛选、旧图切换、末尾图片加载及移动端横向浏览均通过；3L调参后再次检查全部输出尺寸与混合成功/失败案例的筛选。
+
+## 验收重点
+
+- 处理后的9、152号点X差不超过1像素，保证9在152上方。
+- 标准裁切成功时，画面比例及两点位置接近参考构图，保留实际像素，不放大小图。
+- 空间不足的案例明确提示，仍完成扶正。
+- 旧/新规则叠图、参考框与底图坐标一致，原图坐标可通过逆变换还原。
+- 全部30案例均可查看，成功案例显示男女各7型，失败案例显示原图与原因；沿用每人一行、单张旧规则图的展示。
+
+## 2026-11-04 提交前 Review 与验证
+
+- 接口5的请求处理函数、三档PNG生成函数与`a3fddbb`逐项AST比较一致，测试页逐字节一致；新高度规则、扶正裁切和矩形叠加均在接口52。
+- 范围修正：接口52仅属于GPU worker，返回`*_base64`及null；网关路由、落盘、字段改名、mock worker均恢复`a3fddbb`，没有“仅52号改名”的特殊逻辑。测试确认旧网关行为保持原样且未暴露52号路由。
+- 范围修正后的回归检查：`./venv/bin/pytest tests/test_ip_access.py tests/test_gateway_hairline_routes.py tests/test_gateway_nullable_images.py tests/test_api.py -q`，54项通过。
+- GPU worker所有动态接口启用直连IP白名单，本机和网关`49.232.51.171`可访问；网关侧不启用IP限制。配置与部署见[接口IP白名单](接口IP白名单.md)。
+- 上一轮`./venv/bin/pytest -q`：113项通过、1项已有失败（撤回网关新增功能前的测试数）。`tests/test_pipeline.py::test_numeric_regression`期望全脸总高24.01cm，当前实测22.47cm；从未修改的`a3fddbb`导出代码、夹具后独立运行，同样失败。此次保留历史断言，未通过修改基线掩盖差异。
+- 浏览器实测：接口5返回middle/high/low且保持868×1166；接口52长图案例7返回1062×1416、`status=aspect_crop`，四栏图和裁切边框正常；对比页显示30个案例、378组结果，旧书签可跳转52，未出现页面脚本错误。定位验证均关闭ComfyUI生发，本轮未重复运行耗时的真实生发。
+- 批量结果与上传图位于被Git忽略的`static/annotations/`及`tests/output/`，不随提交分发；新环境执行`scripts/compare_hairline_height.py`重建。420是计划组合数，378为当前成功结果，失败组及原因保留在报告中。
@@ -0,0 +1,29 @@
+# GPU worker 接口 IP白名单
+
+仅GPU worker使用项目根目录的`ip_access.json`，不存在时仅允许本机。网关不启用IP限制。配置每次请求重新读取，修改后立即生效。现有`X-Internal-Token`鉴权继续有效。
+
+```json
+{
+  "allowed_ips": ["49.232.51.171"]
+}
+```
+
+当前已将网关IP `49.232.51.171`加入白名单；本机回环地址和网卡自身地址自动允许，其余来源返回HTTP403、`code=1010`。支持IPv4、IPv6及CIDR，不自动允许整个内网。
+
+worker的所有动态路径统一保护，包括5号、52号、其他业务接口、调试接口、代理接口、健康检查和文档接口。`/static/`下的测试页和图片可正常访问，但页面发出的API请求仍需通过IP限制。
+
+启动GPU worker时使用`uvicorn ... --no-proxy-headers`，保留真实连接来源。`run_worker.sh`与`hair-worker.service`已配置该选项。来源被服务器提前改写、地址格式错误或白名单配置无效时拒绝访问。
+
+判断依据是直接连接worker的IP，忽略`X-Forwarded-For`、`X-Real-IP`和`Forwarded`，这些请求头不能用于绕过限制。由网关转发的请求按网关IP判断，网关后面的最终用户无需加入worker白名单。网关保持原有鉴权和转发流程，无需部署此配置。
+
+`ip_access.json`不入Git；配置格式参考`ip_access.example.json`。本机网卡地址在请求时自动获取。
+
+## 部署与维护
+
+首次部署复制`ip_access.example.json`为`ip_access.json`，示例已包含网关`49.232.51.171`；已有配置不覆盖。该操作仅在GPU worker执行，网关服务和启动参数保持原样。
+
+修改`allowed_ips`后保存为有效JSON即可生效，无需重启；建议写入同目录临时文件再原子替换，避免编辑过程中出现半份JSON。文件不存在时仅允许本机，内容损坏时动态接口一律拒绝。更新程序时保留本机配置文件。
+
+IP检查先于Token检查。白名单内请求仍需原来的`X-Internal-Token`；`/health`和`/api/v1/debug/*`等原本免Token的路径仍需通过IP检查。静态页可打开不代表该来源可以调用API。
+
+验证命令：`./venv/bin/pytest tests/test_ip_access.py -q`，覆盖本机、精确白名单、CIDR、IPv6、转发头伪造、热更新、全接口拦截与网关不受限制。
@@ -20,11 +20,14 @@
 | 3 B 端生发 | POST | `/api/v1/hair/grow-b` |
 | 4 用户特征 | POST | `/api/v1/face/features` |
 | 5 发际线 PNG 生成 | POST | `/api/v1/hairline/generate` |
+| 52 发际线新规则与裁切 | POST | `/api/v1/hairline/generate-52` |

 ---

 ## 通用约定

+GPU worker的所有动态接口均启用IP限制：允许本机及`ip_access.json`白名单（当前为网关IP `49.232.51.171`），其他来源返回HTTP403、code=1010；静态页面与图片除外。按直接连接worker的IP判断，网关不启用此限制，已有鉴权继续有效。
+
 以下约定适用于全部 5 个接口（最终以联调为准）。

 ### 请求
@@ -35,7 +38,7 @@
 | 请求方式 | `POST` |
 | 图片传参方式 | 三选一，三种都支持：① `multipart/form-data` 上传文件、② 图片 URL、③ base64 字符串 |
 | 字符编码 | UTF-8 |
-| 坐标基准 | 所有返回的坐标均以**原图像素**为基准（原点为图片左上角，x 向右、y 向下） |
+| 坐标基准 | 默认以**原图像素**为基准；接口52开启预处理时使用响应中的处理后底图坐标（原点为左上角，x向右、y向下） |

 图片传参字段（三者**严格互斥**，必须且只能传其一；传了 0 个或 ≥2 个均报错）：

@@ -81,12 +84,13 @@
 | 1001 | 无法识别人像 | 图片中未检测到人脸 |
 | 1002 | 人像分辨率过低 | 低于最低分辨率要求 |
 | 1003 | 角度问题，非正面照 | 非正面 / 角度过大 |
-| 1004 | gender 必填/非法 | 接口2/5 的 `gender` 缺失或非 `male`/`female` |
+| 1004 | gender 必填/非法 | 接口2/5/52/7 的 `gender` 缺失或非 `male`/`female` |
 | 1005 | 检测到多张人脸 | 默认仅支持单人，检测到 2 人或以上时返回 |
 | 1006 | 文件超出大小限制 | 单文件超过 1 MB |
 | 1007 | 图片参数错误 | file / url / base64 未传，或同时传了多个（三者严格互斥） |
 | 1008 | 图片格式不支持 | 非 JPG / PNG |
 | 1009 | 未授权 | 缺少或错误的 `X-Internal-Token`（`/api/*` 路径鉴权） |
+| 1010 | 来源IP不允许 | 仅GPU worker：直连IP不在本机/白名单内，或来源/白名单配置无效；HTTP403 |

 ---

@@ -109,6 +113,7 @@
 | four_courts | object | 四庭数据，见下表 |
 | seven_eyes | object | 七眼数据，见下表 |
 | landmarks | object | 关键分界点坐标（头顶 / 发际线 / 眉心 / 鼻翼下缘 / 下巴尖），原图像素坐标 |
+| vertical_lines | object | 标注图竖线坐标，见下表。与标注 PNG 上的竖线同一套几何 |
 | left_position | object | MediaPipe 21 号关键点坐标（左脸定位点），原图像素：`{ "x": int, "y": int }` |
 | right_position | object | MediaPipe 251 号关键点坐标（右脸定位点，与 21 号镜像），原图像素：`{ "x": int, "y": int }` |

@@ -138,7 +143,28 @@
 | eye6 | number | 从左到右第 6 段宽度（cm）：右眼外角 → 右脸颊 |
 | eye7 | number \| null | 从左到右第 7 段宽度（cm）：右脸颊 → 人头最右（右耳外侧段）。该侧耳朵不可见时为 null |

-> `eye1`~`eye7` 为从左到右共 7 段宽度，与标注图竖线一一对应。最左/最右端线取自耳朵分割外缘；某侧耳朵被头发或侧脸遮挡（不可见）时该侧端线省略，对应 `eye1` 或 `eye7` 为 `null`（键始终保留），实际有效段为 5 或 6 段。`eye3`/`eye5` 为左右眼宽、`eye4` 为两眼间距，与 `eye_width_cm` / `inter_eye_distance_cm` 语义一致。
+> `eye1`~`eye7` 为从左到右共 7 段宽度，与标注图竖线一一对应。最左/最右端线取自耳朵分割外缘；某侧耳朵被头发或侧脸遮挡（不可见）时该侧端线省略，对应 `eye1` 或 `eye7` 为 `null`（键始终保留），实际有效段为 5 或 6 段。`eye3`/`eye5` 为左右眼宽、`eye4` 为两眼间距，与 `eye_width_cm` / `inter_eye_distance_cm` 语义一致。`eye1`/`eye7` 的厘米数由 `vertical_lines` 里对应端点的 x 相减得到。
+
+`vertical_lines`（竖线，原图像素，原点左上，x 向右，y 向下）：
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| y_top | int | 竖线共用上端。接口 1 在头顶之上，伸出 `max(6, round(短边 × 0.03))` 像素，可能略小于 0 |
+| y_bottom | int | 竖线共用下端。接口 1 在下巴尖之下，伸出同样像素，可能略大于图高 |
+| lines | array | 从左到右实际画出的竖线。每项 `{ "name": string, "x": int }` |
+
+`lines[].name` 从左到右：
+
+| name | 竖线 |
+|------|------|
+| left_head_edge | 人头最左（耳朵外缘）。该侧耳朵不可见时不出现，此时 `eye1` 为 `null` |
+| left_cheek | 左脸颊 |
+| left_eye_outer | 左眼外角 |
+| left_eye_inner | 左眼内角 |
+| right_eye_inner | 右眼内角 |
+| right_eye_outer | 右眼外角 |
+| right_cheek | 右脸颊 |
+| right_head_edge | 人头最右（耳朵外缘）。该侧耳朵不可见时不出现，此时 `eye7` 为 `null` |

 ### 标注图片（UI）规范

@@ -182,6 +208,19 @@
      "brow_center": { "x": 540, "y": 740 },
      "nose_bottom": { "x": 540, "y": 1050 },
      "chin_tip": { "x": 540, "y": 1360 }
+    },
+    "vertical_lines": {
+      "y_top": 98, "y_bottom": 1410,
+      "lines": [
+        { "name": "left_head_edge", "x": 120 },
+        { "name": "left_cheek", "x": 210 },
+        { "name": "left_eye_outer", "x": 310 },
+        { "name": "left_eye_inner", "x": 420 },
+        { "name": "right_eye_inner", "x": 560 },
+        { "name": "right_eye_outer", "x": 670 },
+        { "name": "right_cheek", "x": 780 },
+        { "name": "right_head_edge", "x": 890 }
+      ]
    }
  }
 }
@@ -213,6 +252,7 @@
 | four_courts | object | 三庭数据（上/中/下庭，各含 cm 与 ratio；**无顶庭**） |
 | seven_eyes | object | 七眼数据（眼宽/脸宽/两眼间距 cm + 占比 ratios + **eye2~eye6** 共 5 段宽度） |
 | landmarks | object | 四个关键点像素坐标（发际线/眉心/鼻翼下缘/下巴尖） |
+| vertical_lines | object | 标注图竖线坐标。结构同接口 1，但 `y_top`/`y_bottom` 为发际线到下巴尖（不伸出），`lines` 只有中间 6 条，无 `left_head_edge` / `right_head_edge` |
 | left_position | object | MediaPipe 21 号关键点坐标（左脸定位点），原图像素：`{ "x": int, "y": int }` |
 | right_position | object | MediaPipe 251 号关键点坐标（右脸定位点，与 21 号镜像），原图像素：`{ "x": int, "y": int }` |

@@ -242,6 +282,17 @@
      "brow_center": { "x": 540, "y": 740 },
      "nose_bottom": { "x": 540, "y": 1050 },
      "chin_tip": { "x": 540, "y": 1360 }
+    },
+    "vertical_lines": {
+      "y_top": 430, "y_bottom": 1360,
+      "lines": [
+        { "name": "left_cheek", "x": 210 },
+        { "name": "left_eye_outer", "x": 310 },
+        { "name": "left_eye_inner", "x": 420 },
+        { "name": "right_eye_inner", "x": 560 },
+        { "name": "right_eye_outer", "x": 670 },
+        { "name": "right_cheek", "x": 780 }
+      ]
    }
  }
 }
@@ -449,6 +500,7 @@
 | four_courts | object | 四庭数据（顶/上/中/下庭 cm + 占比 ratios），结构同接口1 |
 | seven_eyes | object | 七眼数据（眼宽/脸宽/两眼间距 cm + 占比 ratios + eye1~eye7 从左到右 7 段宽度），结构同接口1 |
 | landmarks | object | 5 个纵向关键点像素坐标（hair_top/hairline/brow_center/nose_bottom/chin_tip），结构同接口1 |
+| vertical_lines | object | 竖线坐标，结构同接口1 |
 | hairline_source | string | 发际线来源：`segmentation`（真实分割）/ `estimated`（比例估算）/ `corrected`（发际线贴近头顶、分割不可靠，已按 0.95×中庭反推修正） |
 | head_pose | object | 头部姿态角度（yaw/pitch/roll，单位：度） |
 | left_position | object | MediaPipe 21 号关键点坐标（左脸定位点），原图像素：`{ "x": int, "y": int }` |
@@ -518,6 +570,155 @@

 ---

+## 接口 52：发际线 PNG 生成接口
+
+**说明**：入参同接口2（先选性别、再多选发型）。对每个选中发型返回一张 `middle` 设计叠图与生发图。**目标上庭 =（中庭 + 下庭）÷2**，从眉心向上定位；所有形状的中点对齐同一目标。`high` / `low` 已停用，旧字段保留为 `null`。
+
+**请求**：`POST /api/v1/hairline/generate-52`（仅GPU worker，端口8187；本次不在网关新增路由）
+
+**IP限制**：GPU worker的所有动态接口统一仅允许本机回环地址、本机网卡地址及白名单；目前白名单含网关IP `49.232.51.171`，其他直连来源需加入`ip_access.json`的`allowed_ips`。该文件支持单个IP与CIDR，修改后立即生效；拒绝时返回HTTP403、code=1010。网关不启用此限制，已有鉴权仍保留。详见[接口IP白名单](接口IP白名单.md)。
+
+### 图片扶正与标准构图
+
+显式传`normalize_image=true`时，先扶正9→152，再按**高(3÷0.9)×L≈3.3333×L、宽2.5×L（3:4）**裁切，L为9→152的直线像素距离。相对原3L规则，裁切框宽高均扩大约11.11%，同样显示尺寸下人物缩小约10%。高度取最接近的4的倍数，宽度取高度的3/4。眉心定位沿用已扶正参考图：9号点相对位置约(49.0061%,34.4136%)。参考图仍为786×1048，两点X均为385.18784。
+
+目标窗口完全在原图有效区域内才进行标准裁切。标准窗口不足时，偏长图片（宽/高<3/4）采用原图有效范围内最大的3:4窗口，尽量贴近眉心目标位置并保留完整人脸，`status=aspect_crop`；人脸放不进3:4窗口时回退仅扶正并返回原因。其余图片空间不足时仍仅扶正、去除旋转空角。不放大小图。**所有叠图与坐标使用响应中的处理后底图 `input_image_*`**，调用方应使用该底图叠加；`normalize_image=false` 可关闭此预处理。
+
+`framing`包含处理前后尺寸、旋转角、是否标准裁切、跳过原因、`source_to_processed`/`processed_to_source`仿射矩阵。`status`区分`reference_crop`（标准窗口）、`aspect_crop`（长图3:4裁切）和`rotation_only`（仅扶正）；`standard_crop=false`时仍可能有`aspect_crop=true`。`aspect_crop_skip_reason`说明长图无法保留人脸时的回退原因。`original_image_*`保留EXIF方向修正后的上传图，原图坐标可通过逆矩阵还原。
+
+### 输入
+
+图片参数见「通用约定 → 图片传参字段」。专属参数：
+
+| 参数 | 类型 | 必填 | 说明 |
+|------|------|------|------|
+| gender | string | **是** | 性别：`male` / `female`。决定发型集合（female 7 / male 7）。缺失/非法返回 `1004` |
+| hair_style | string | **是** | 发型序号，**逗号分隔多选**（如 `1,2,3`），决定返回哪些发际线类型。female：1=ellipse, 2=flower, 3=heart, 4=straight, 5=wave, 6=bigflower, 7=clasicalflower；male：1=ellipse, 2=inverse_arc, 3=m, 4=straight, 5=heart, 6=Softpetal, 7=wave。缺失/越界/非法返回 `1007` |
+| use_mask | bool | 否 | 生发是否启用 inpaint 遮罩，默认 `true`。`false` 时用干净原图生成（空遮罩、不烧模板黑线），供测试对比 |
+| prompt | string | 否 | ComfyUI 提示词，默认「填充遮罩区域的头发，人物增加一点磨皮」，会替换工作流节点 60 的文本 |
+| normalize_image | bool | 否 | 新增可选开关，默认false：保留原图尺寸与坐标。传true开启扶正和裁切；规则测试页显式传true，对比调试页默认开启 |
+| generate_grow_image | bool | 否 | 是否生成生发效果图（ComfyUI 生发，全流程最耗时），默认 `true`。传 `false` 时跳过生发，各发型 `grown_image_*` 恒为 `null`，仅返回设计叠图、中心点与高度规则数据，可大幅降低耗时 |
+
+> 生发划线与遮罩直接取新位置的 middle 曲线；旧规则的头发边界与模板用于保留形状和倾斜，不决定新中心高度。计算使用未取整的像素高度，与四庭测量采用同一 Y 方向；测量主流程失败时用发际线管线的同编号关键点计算，`measurement_source` 标明来源。目标越出图片或无法对齐时返回 1007，不静默回退旧规则。
+
+### 高度规则与对比
+
+- **参数与字段**：原有8个请求参数、响应外层结构和所有旧返回字段均保留。`use_mask=true`、默认prompt、`generate_grow_image=true`保持原值。新功能只新增可选参数和返回字段；旧请求未传`normalize_image`时保留原图坐标，high/low按已约定保留null，worker始终返回`*_base64`，不负责保存图片、生成公网URL或改名；网关保持原样。
+- `height_rule`：`middle_height_px`、`lower_height_px`、`target_upper_height_px`、`target_upper_height_cm`（可空）、`target_center`、`brow_center`、`nose_bottom`、`chin_tip`、`measurement_source`。
+- 每项 `height_placement`：`old_center`、`new_center`、`shift_x_px`、`shift_y_px`（正=下移）、`target_error_px`。浮点坐标保留精度，旧中心点字段继续返回取整像素。
+- `face_measure` 为处理后底图的实际测量值，未被设计目标覆盖。
+- `guide_landmarks`：9、150、152、162、251、389 号 MediaPipe 点的图片像素坐标，供对比页绘制参考框。参考框左/右边分别经过162/389，底边经过9，高度为9→152距离的70%；两侧均固定平行9→152。
+- 规则页：`/static/test_interface52.html`；旧/新规则对比：`/static/test_interface52_compare.html`。
+- 规则页每个发型依次展示未裁切原图、middle、裁切发际线叠加、生发图。“裁切发际线叠加”由页面按`guide_landmarks`生成：在相同处理后底图上叠加矩形内的middle PNG，矩形外保持透明；左/右边过162/389，底边过9，高度0.7×9→152，左右边平行9→152。
+- 对比调试接口：`POST /api/v1/debug/hairline-height-compare`，图片三选一、gender、hair_style；默认只出旧/新曲线，可选 `generate_grow_image=true` 额外验证新规则生发。
+- 测试清单：`tests/fixtures/hairline_height_cases.json`（用户指定30张）；批量运行：`./venv/bin/python scripts/compare_hairline_height.py`，默认全部30张、男女各7型，共420组。每人一行，未裁切原图放在最前，后接一张可切换样式的旧规则图和全部新规则样式。可用 `--cases 1,2,3,4,5` 指定部分案例。
+
+### 输出（data）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| hairline_images | object[] | **选中发型**列表，**数量 = 所选发型数**，元素见下表 |
+| best_hairline_center_point | object \| null | **首个选中发型**的 **middle 档**发际线曲线「面部中间点」坐标，当前底图像素：`{ "x": number, "y": number }` |
+| high_hairline_center_point | object \| null | 已停用，固定 `null` |
+| low_hairline_center_point | object \| null | 已停用，固定 `null` |
+| input_image_base64 | string | 当前底图的JPEG base64，发际线叠图、参考点和测量值均使用该图的坐标系 |
+| original_image_base64 | string | EXIF方向修正后的上传原图JPEG base64 |
+| width / height | int | 当前底图的实际像素宽 / 高 |
+| coordinate_space | string | 开启预处理时为 `processed_image`，关闭时为 `original_image` |
+| framing | object \| null | 裁切状态、旋转角、前后尺寸、轴线误差及双向仿射矩阵；关闭预处理时为 `null` |
+| height_rule | object | 中庭/下庭、目标上庭高度、目标中心及测量来源，详见“高度规则与对比” |
+| guide_landmarks | object | 9、150、152、162、251、389 号点的当前底图像素坐标；参考框实际使用9、152、162、389 |
+| face_measure | object \| null | **复用接口1**的四庭七眼测量**数值**（不含标注图）。独立流程，测量失败（无人脸/非正面/分割失败）时为 `null`，不影响发际线主结果。字段结构见下表 |
+
+`hairline_images` 元素：
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| hairline_type | string | 发际线类型 key：`ellipse`/`flower`/`heart`/`straight`/`wave`/`bigflower`/`clasicalflower`（female），`ellipse`/`m`/`straight`/`inverse_arc`/`heart`/`Softpetal`/`wave`（male） |
+| image_middle_base64 | string | middle 档发际线曲线**透明 PNG** base64（仅曲线，透明底，**不含人物**，需叠加 `input_image_base64` 显示） |
+| image_high_base64 | null | 已停用，保留字段为 `null` |
+| image_low_base64 | null | 已停用，保留字段为 `null` |
+| grown_image_base64 | string \| null | **生发后图片** JPEG base64（ComfyUI「植发」效果图，完整人像照片，生发失败或 `generate_grow_image=false` 时为 `null`） |
+| order | int | 发型序号（= 传入的 hair_style 值） |
+| height_placement | object | 旧/新中心、XY移动距离、目标误差，均为未取整的像素值 |
+
+> 上表为GPU worker的实际响应字段；所有图片使用base64，空图片保留原字段并返回null。网关的落盘与URL改写属于网关职责，本次不修改。
+>
+> 前端将base64转为data URI后显示。middle透明图叠加到`input_image_base64`，生发图直接显示。GPU worker本机测试页为`http://localhost:8187/static/test_interface52.html`。
+
+`face_measure` 元素（与[接口1](#接口-1四庭七眼测量标注接口)的 `data` 同构，**不含** `annotated_image_*` 标注图字段）：
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| face_total_height_cm | number | 全脸总高度（cm）= 四庭之和 |
+| four_courts | object | 四庭数据（顶/上/中/下庭 cm + 占比 ratios），结构同接口1 |
+| seven_eyes | object | 七眼数据（眼宽/脸宽/两眼间距 cm + 占比 ratios + eye1~eye7 从左到右 7 段宽度），结构同接口1 |
+| landmarks | object | 5 个纵向关键点像素坐标（hair_top/hairline/brow_center/nose_bottom/chin_tip），结构同接口1 |
+| vertical_lines | object | 竖线坐标，结构同接口1 |
+| hairline_source | string | 发际线来源：`segmentation`（真实分割）/ `estimated`（比例估算）/ `corrected`（发际线贴近头顶、分割不可靠，已按 0.95×中庭反推修正） |
+| head_pose | object | 头部姿态角度（yaw/pitch/roll，单位：度） |
+| left_position | object | MediaPipe 21 号关键点坐标（左脸定位点），当前底图像素：`{ "x": int, "y": int }` |
+| right_position | object | MediaPipe 251 号关键点坐标（右脸定位点，与 21 号镜像），当前底图像素：`{ "x": int, "y": int }` |
+
+> `eye1`~`eye7` 为从左到右共 7 段宽度，eye1=左耳外段、eye7=右耳外段，某侧耳朵不可见时对应段为 `null`。详见接口1说明。
+
+### GPU worker响应结构示例（normalize_image=false）
+
+此例未开启裁切，坐标基于原图；`face_measure=null`时仍可按发际线关键点计算目标。图片内容以缩略base64字符串示意，空图片字段返回null，字段名不变。
+
+```json
+{
+  "code": 0,
+  "message": "success",
+  "request_id": "mock-request-id",
+  "data": {
+    "hairline_images": [
+      {
+        "hairline_type": "ellipse",
+        "image_middle_base64": "iVBORw0KGgo...",
+        "image_high_base64": null,
+        "image_low_base64": null,
+        "grown_image_base64": "/9j/4AAQ...",
+        "order": 1,
+        "height_placement": {
+          "old_center": {"x": 540.0, "y": 450.0},
+          "new_center": {"x": 540.0, "y": 430.0},
+          "shift_x_px": 0.0, "shift_y_px": -20.0, "target_error_px": 0.0
+        }
+      }
+    ],
+    "best_hairline_center_point": {"x": 540, "y": 430},
+    "high_hairline_center_point": null,
+    "low_hairline_center_point": null,
+    "input_image_base64": "/9j/4AAQ...",
+    "original_image_base64": "/9j/4AAQ...",
+    "width": 1080, "height": 1440,
+    "coordinate_space": "original_image",
+    "framing": null,
+    "height_rule": {
+      "name": "court_average", "formula": "上庭 = (中庭 + 下庭) / 2",
+      "measurement_source": "hairline_landmarks",
+      "brow_center": {"x": 540.0, "y": 740.0},
+      "nose_bottom": {"x": 540.0, "y": 1050.0},
+      "chin_tip": {"x": 540.0, "y": 1360.0},
+      "middle_height_px": 310.0, "lower_height_px": 310.0,
+      "target_upper_height_px": 310.0, "target_upper_height_cm": null,
+      "target_center": {"x": 540.0, "y": 430.0}
+    },
+    "guide_landmarks": {
+      "9": {"x": 540.0, "y": 740.0}, "150": {"x": 470.0, "y": 1340.0},
+      "152": {"x": 540.0, "y": 1360.0}, "162": {"x": 300.0, "y": 680.0},
+      "251": {"x": 770.0, "y": 630.0}, "389": {"x": 780.0, "y": 680.0}
+    },
+    "face_measure": null
+  }
+}
+```
+
+开启`normalize_image=true`后，`coordinate_space=processed_image`，`framing`包含前述裁切状态和坐标变换，所有发际线图片须叠加到`input_image_base64`。原始图片仅用于展示和坐标回溯。
+
+---
+
 ## 接口 9：头发遮罩生成接口

 **说明**：输入一张含人头的照片，生成头发遮罩，并返回**每一步的可视化图**（供对比调试）。
@@ -702,6 +903,7 @@
 | 3 B 端生发 | `/api/v1/hair/grow-b` | 划线图片 | 检测到的发际线 + 生发后图片 |
 | 4 用户特征 | `/api/v1/face/features` | 用户照片 | 6 个用户特征字段（脸形/眉形/年龄/动静/性别/基因风格），网关本机实现 |
 | 5 发际线 PNG | `/api/v1/hairline/generate` | 用户照片 + gender + hair_style（多选） | 每个选中发型 middle/high/low 三档发际线叠图 + 生发图 + 最合适发际线面部中间点坐标 |
+| 52 发际线新规则 | `/api/v1/hairline/generate-52` | 用户照片 + gender + hair_style（多选） | 每个选中发型的新规则 middle 设计叠图 + 生发图 + 目标中心坐标；high/low 旧字段为 null |
 | 9 头发遮罩 | `/api/v1/head/mask` | 用户照片 | 头发遮罩 + 分步可视化（BiSeNet/SegFormer 双路） |
 | 10 外缘膨胀带 | `/api/v1/head/band` | 用户照片 | 头部外缘带状遮罩 + 分步可视化 |
 | 11 发际线生发 | `/api/v1/hairline/grow` | 用户照片 + hairline_id | 压低发际线图 + 全步骤可视化（不含重绘） |
@@ -716,7 +918,7 @@

 - 图片传参：**文件 / URL / base64 三种都支持，严格互斥**（必须且只能传一个，base64 需带 `data:image/...;base64,` 前缀）。
 - 默认单人；检测到 2~3 人报错提示客户端；单文件 **≤ 1 MB**；格式仅 **JPG / PNG**。
- 所有坐标基准：**原图像素**。
+- 坐标基准：默认**原图像素**；接口52开启预处理时使用响应中的**处理后底图像素**。
 - 接口 2：提供「美颜」开关 `beauty_enabled`。
 - 接口 3：仅传**划线图**（`marked_image_*` 三选一），需返回发际线类型（`custom`）。
 - 接口 4：`data` 返回**一个 JSON 字符串**，字段不固定，可随时调整。
@@ -0,0 +1,234 @@
+# 网关摘除隔离期实施方案（方案 B）
+
+> 目标读者：网关机器的部署/维护者。本文档给出基于 **main 分支** `gateway/` 代码的
+> 完整改动说明，可直接照做。**不需要改 worker 侧任何代码。**
+
+## 1. 背景与事故复盘
+
+2026-09-19 晚，GPU worker 机器（ubuntu3090_debug_4090 所在主机）磁盘写满，业务接口
+全部失败，但网关**持续把请求打到该机器近 16 个小时**（21:30 故障发生 → 次日 13:36
+人工重启），没有切换到其他 worker。复盘结论是三层防线同时失效：
+
+| 层 | 问题 | 后果 |
+|---|---|---|
+| 1. worker `/health` 太浅 | 只返回内存里的"模型就绪"标志，磁盘满/下游 ComfyUI 挂掉不影响它 | 探测一直 200 |
+| 2. 业务错误不触发摘除 | 下游出错经包装后以 HTTP 200/400 + 错误码返回，`forward.py` 只对 连接异常/超时/≥500/401 摘除 | 请求失败但 worker 被视为健康 |
+| 3. 摘除后 8 秒即复活（**本方案要修的**） | `mark_worker_unhealthy()` 下线后，下一轮探测（8s 后）`/health` 返回 200，`healthy_threshold=1` 使其**一次成功立即重新上线** | 即使被摘除也会反复横跳，流量持续打入 |
+
+worker 侧的第 1 层已经修复（方案 A：`/health` 深检磁盘水位 + 下游依赖 TCP 探测，
+异常返回 503，代码在分支 `ubuntu3090_debug_4090`，提交 `14562d4`）。本方案修复
+第 3 层：**被动摘除后进入隔离期，隔离期内即使探测成功也不复活**。两层配合后：
+worker 磁盘满 → `/health` 503 → 网关 2 次探测失败摘除 → 隔离期内不再接流量。
+
+> 第 2 层（错误码语义化，涉及调用方兼容）暂不处理，单独立项。
+
+## 2. 现状代码行为（main 分支 gateway/pool.py）
+
+- `mark_worker_unhealthy(w)`：被 `forward.py` 在转发失败（连接异常/超时/5xx/401）时
+  调用，置 `w.online = False`。**没有任何冷却机制**。
+- `_health_check_loop(cfg)`：每 `interval_seconds`（默认 8s）探测所有 worker 的
+  `/health`；返回 200 时 `consecutive_successes += 1`，达到 `healthy_threshold`
+  （默认 **1**）且当前 offline → 立即 `online = True`。
+- 因此被动摘除最多 8 秒后就被一次探测成功拉回——如果 worker 是"浅健康"或故障
+  间歇性表现，就会陷入 摘除→复活→失败→摘除 的横跳。
+- 另外 58-96 行有一段**从未被调用的死代码** `_check_worker_health()` /
+  `_mark_failure()`（写了一半的旧实现），本次顺带删除，避免后人改错地方。
+
+## 3. 代码改动（共 2 个文件 + 1 个配置）
+
+### 3.1 `gateway/pool.py`
+
+**(1) `WorkerState` 增加一个字段：**
+
+```python
+@dataclass
+class WorkerState:
+    """单个 worker 的运行时状态。"""
+    url: str
+    online: bool = False  # 初始 offline，等健康检查通过后上线
+    busy: bool = False
+    consecutive_failures: int = 0
+    consecutive_successes: int = 0
+    offline_until: float = 0.0  # 被动摘除后的隔离截止时刻（time.monotonic() 基准）
+```
+
+**(2) 模块顶部（`logger = ...` 之后）加一个配置读取 helper：**
+
+```python
+def _cooldown_seconds() -> float:
+    """被动摘除后的隔离期时长（秒），配置 health_check.cooldown_seconds，默认 60。"""
+    try:
+        from gateway.config import get_config
+        return float(get_config()["health_check"].get("cooldown_seconds", 60))
+    except Exception:  # noqa: BLE001
+        return 60.0
+```
+
+（函数内 import 是为了不引入模块级循环依赖；`gateway/config.py` 不依赖 pool，顶层
+import 也可以。）
+
+**(3) `mark_worker_unhealthy()` 摘除时同时设置隔离期：**
+
+```python
+async def mark_worker_unhealthy(w: WorkerState) -> None:
+    """被动标记：转发失败时立即将该 worker 下线，并进入隔离期。
+
+    隔离期内即使 /health 探测成功也不复活（由 _health_check_loop 守卫），
+    避免"摘除→8 秒复活→再失败"的横跳。
+    """
+    async with _pool_condition:
+        if w.online:
+            w.online = False
+            w.consecutive_failures += 1
+            w.offline_until = time.monotonic() + _cooldown_seconds()
+            logger.warning("⚠ Worker 被动下线: %s（转发失败，隔离 %.0fs）",
+                           w.url, _cooldown_seconds())
+```
+
+**(4) `_health_check_loop()` 的探测成功分支加隔离期守卫。** 找到这段：
+
+```python
+                if resp.status_code == 200:
+                    w.consecutive_failures = 0
+                    w.consecutive_successes += 1
+                    if w.consecutive_successes >= healthy_threshold and not w.online:
+                        w.online = True
+                        logger.info("✅ Worker 上线: %s", w.url)
+```
+
+改为：
+
+```python
+                if resp.status_code == 200:
+                    w.consecutive_failures = 0
+                    w.consecutive_successes += 1
+                    if w.consecutive_successes >= healthy_threshold and not w.online:
+                        if time.monotonic() < w.offline_until:
+                            logger.debug("Worker %s 探测成功但处于隔离期，暂不复活", w.url)
+                        else:
+                            w.online = True
+                            logger.info("✅ Worker 上线: %s", w.url)
+```
+
+注意：`except`（连接异常）分支只累计失败、不涉及复活，无需改动；
+`init_pool()` 启动时的首轮探测同理无需改动（进程刚启动时 `offline_until` 为 0，
+不构成隔离）。
+
+**(5) 删除死代码**：`_check_worker_health()` 与
+`_mark_failure()` 两个函数整体删除（约 58-96 行，以函数名为准，别删到
+`_health_check_loop`）。删除前可确认无调用方：
+`grep -rn "_check_worker_health\|_mark_failure" .` 应只剩定义处。
+
+### 3.2 `gateway/config.py`
+
+`DEFAULTS["health_check"]` 增加一个 key：
+
+```python
+    "health_check": {
+        "path": "/health",
+        "interval_seconds": 8,
+        "timeout_seconds": 3,
+        "unhealthy_threshold": 2,
+        "healthy_threshold": 1,
+        "cooldown_seconds": 60,   # ← 新增：被动摘除后的隔离期（秒）
+    },
+```
+
+`config.example.json` 同步加这个 key。网关机器的实际 `config.json` **不加也不会报错**
+（代码里有 60 的默认兜底），但建议显式加上，便于运维感知。
+
+### 3.3 建议单测：`tests/test_gateway_pool.py`（新文件）
+
+gateway 子包目前没有任何测试。隔离期是纯逻辑，建议至少覆盖以下三个用例
+（不依赖真实网络）：
+
+```python
+"""gateway/pool.py 隔离期逻辑单测。"""
+import asyncio
+import time
+from unittest.mock import patch
+
+from gateway import pool
+from gateway.pool import WorkerState
+
+
+def _drain(coro):
+    return asyncio.run(coro)
+
+
+def test_passive_ejection_sets_cooldown():
+    w = WorkerState(url="http://127.0.0.1:9000", online=True)
+    async def run():
+        pool._pool_condition = asyncio.Condition()
+        with patch.object(pool, "_cooldown_seconds", return_value=60.0):
+            await pool.mark_worker_unhealthy(w)
+    _drain(run())
+    assert w.online is False
+    assert w.offline_until > time.monotonic()
+
+
+def test_probe_success_revives_after_cooldown():
+    w = WorkerState(url="http://127.0.0.1:9000")  # offline
+    w.offline_until = time.monotonic() - 1        # 隔离期已过
+    # 直接验证状态迁移逻辑（若按 3.1(4) 内联实现，可复制该分支逻辑做断言，
+    # 或将判断抽成 _try_revive(w, healthy_threshold) 小函数后单测）
+    ...
+
+
+def test_probe_success_blocked_during_cooldown():
+    w = WorkerState(url="http://127.0.0.1:9000")
+    w.offline_until = time.monotonic() + 60
+    ...  # 断言保持 offline
+```
+
+> 提示：如果按 3.1(4) 原样内联修改，后两个用例不好直接驱动，可以把"是否允许复活"
+> 抽成一个 6 行的纯函数 `_try_revive(w, healthy_threshold) -> bool` 再测，主循环
+> 调用它即可——实现方式自定，覆盖语义即可。
+
+## 4. 部署步骤（网关机器）
+
+1. `git pull` main 分支（含本方案代码改动）。
+2. （可选但建议）`config.json` 的 `health_check` 段加 `"cooldown_seconds": 60`。
+3. 重启网关服务（`hair-gateway.service` 或等价的 uvicorn 进程）。重启期间网关
+   本身短暂不可用，选低峰执行。
+4. 观察启动日志：`健康检查循环启动 | 间隔=8s | 下线阈值=2 | 上线阈值=1 | workers=N`
+   正常出现。
+
+## 5. 验证
+
+**功能验证（低峰演练）：**
+
+1. 人工把主 worker 摘除一次最直接的方法：临时在 worker 机器
+   `systemctl stop hair-worker`（或临时设 `HEALTH_MIN_FREE_GB=999` 重启）。
+2. 观察网关日志：
+   - 连续 2 次探测失败 → `⚠ Worker 下线`（原有行为）；
+   - **转发失败触发的** `⚠ Worker 被动下线: ...（转发失败，隔离 60s）`（新日志）；
+   - 恢复 worker 后，若刚被被动摘除过，即使 `/health` 恢复 200，也应看到 60 秒内
+     **不出现** `✅ Worker 上线`，隔离期满后下一次探测成功才上线。
+3. `GET /gateway-health` 观察 `workers_healthy` 的变化节奏。
+
+**回归确认：** 正常场景 worker 上下线不受影响（隔离期只对"被动摘除"生效；
+探测失败导致的下线，恢复后仍按 `healthy_threshold` 立即上线）。
+
+## 6. 回滚
+
+- 代码回滚：`git revert` 对应提交即可，无数据/配置迁移。
+- 紧急关闭隔离期：网关机器 `config.json` 设 `"cooldown_seconds": 0`
+  （等价于旧行为——摘除后下一次探测成功即复活），无需回代码。
+
+## 7. 附：worker 侧方案 A 的行为（已上线，供联调参考）
+
+worker（分支 `ubuntu3090_debug_4090`，提交 `14562d4`）的 `/health` 现在会在以下
+情况返回 **503**（网关 2 次探测失败即摘除）：
+
+- 模型加载中：`503 {"status":"starting"}`（原有行为不变）
+- 磁盘剩余低于阈值：`503 {"status":"degraded","reason":"disk_free_below_5.0g","disk_free_gb":...}`
+- 下游依赖（ComfyUI 8188 / change_hair 8801）TCP 不通：
+  `503 {"status":"degraded","reason":"dep_down:comfyui,...","deps":{...}}`
+
+正常时返回 `200 {"status":"ok","disk_free_gb":24.6,"deps":{"comfyui":true,"change_hair":true}}`。
+
+worker 侧配置（`worker_config.json` 的 `health` 段，环境变量可覆盖）：
+`min_free_gb`（默认 5，0=关闭）、`check_deps`（默认 true）、`dep_comfyui`（默认
+127.0.0.1:8188）、`deps_cache_seconds`（默认 5）。**其他 worker 机器如果也想启用
+深检，cherry-pick 该提交并补一份 worker_config.json 的 health 段即可。**
@@ -180,6 +180,60 @@ def _ear_edges_from_mask(ear_mask, hair_mask, y0, y1, left_cheek_x, right_cheek_
    return head_l, head_r


+# 返回给接口的竖线名 → eyes["points"] 里的键。从左到右的解剖顺序。
+_VERTICAL_INNER_LINES = (
+    ("left_cheek", "left_cheek"),
+    ("left_eye_outer", "left_outer"),
+    ("left_eye_inner", "left_inner"),
+    ("right_eye_inner", "right_inner"),
+    ("right_eye_outer", "right_outer"),
+    ("right_cheek", "right_cheek"),
+)
+
+
+def vertical_line_geometry(measure_result, image_width, image_height,
+                           ear_mask=None, hair_mask=None, variant="v1"):
+    """标注图竖线的像素几何，供画线与接口返回共用。
+
+    返回 {"y_top", "y_bottom", "lines": [{"name", "x"}, ...]}。
+    x / y 均为原图像素整数（四舍五入）。lines 从左到右，只含实际画出的线。
+    variant="v1"：竖线从头顶上方延伸到下巴下方；有耳朵外缘时含
+    left_head_edge / right_head_edge。
+    variant="v6"：竖线从发际线到下巴尖，不含人头两端。
+    """
+    v = measure_result.vertical
+    s = min(image_width, image_height)
+    if variant == "v6":
+        order = ["hairline", "brow_center", "nose_bottom", "chin_tip"]
+    else:
+        order = ["hair_top", "hairline", "brow_center", "nose_bottom", "chin_tip"]
+    ys = [v[name][1] for name in order]
+
+    pts = measure_result.eyes["points"]
+    named = [(public, float(pts[internal][0])) for public, internal in _VERTICAL_INNER_LINES]
+    if variant != "v6":
+        lcx, rcx = pts["left_cheek"][0], pts["right_cheek"][0]
+        head_l, head_r = _ear_edges_from_mask(
+            ear_mask, hair_mask, ys[0], ys[-1], lcx, rcx, (lcx + rcx) / 2)
+        if head_l is not None:
+            named.append(("left_head_edge", float(head_l)))
+        if head_r is not None:
+            named.append(("right_head_edge", float(head_r)))
+    named.sort(key=lambda item: item[1])
+
+    fy0, fy1 = ys[0], ys[-1]
+    over = max(6, round(s * 0.030))
+    if variant == "v6":
+        y_top, y_bottom = fy0, fy1
+    else:
+        y_top, y_bottom = fy0 - over, fy1 + over
+    return {
+        "y_top": int(round(y_top)),
+        "y_bottom": int(round(y_bottom)),
+        "lines": [{"name": name, "x": int(round(x))} for name, x in named],
+    }
+
+
 def create_annotated_image(image_bgr, measure_result, ear_mask=None, hair_mask=None,
                           variant="v1"):
    """生成标注图层 PNG（透明底 RGBA，尺寸同原图）。返回 PIL.Image。
@@ -221,19 +275,11 @@ def create_annotated_image(image_bgr, measure_result, ear_mask=None, hair_mask=N
    ys = [v[name][1] for name in order]

    pts = measure_result.eyes["points"]
-    seven_keys = ["left_cheek", "left_outer", "left_inner",
-                  "right_inner", "right_outer", "right_cheek"]
-    base_xs = [pts[k][0] for k in seven_keys]
-    if variant == "v6":
-        xs = sorted(base_xs)                  # 接口6：仅七眼 6 点，不画人头最左/最右端线
-    else:
-        # 人头最左/最右：取自耳朵分割掩膜外缘（方案B，类7/8），脸纵向范围内统计。
-        # 看不到耳朵（被头发/侧脸遮挡 → 掩膜空）或外缘未越过脸颊线则省略该侧端线。
-        lcx, rcx = pts["left_cheek"][0], pts["right_cheek"][0]
-        head_l, head_r = _ear_edges_from_mask(
-            ear_mask, hair_mask, ys[0], ys[-1], lcx, rcx, (lcx + rcx) / 2)
-        head_xs = [x for x in (head_l, head_r) if x is not None]
-        xs = sorted(base_xs + head_xs)        # 自左向右（6 或 7/8 点）
+    # 竖线 x 与纵向范围与接口返回的 vertical_lines 同一份几何。
+    geom = vertical_line_geometry(
+        measure_result, w, h, ear_mask=ear_mask, hair_mask=hair_mask, variant=variant)
+    xs = [line["x"] for line in geom["lines"]]
+    v_top, v_bot = geom["y_top"], geom["y_bottom"]

    # 人脸/人头包围盒
    fx0, fx1 = xs[0], xs[-1]
@@ -241,9 +287,6 @@ def create_annotated_image(image_bgr, measure_result, ear_mask=None, hair_mask=N
    face_cx = (fx0 + fx1) / 2
    over = max(6, round(s * 0.030))           # 线超出包围盒的长度（参考图风格）
    face_half = (fx1 - fx0) / 2 + over        # 横线超出最外侧竖线一点
-    # 竖线纵向范围：v6 = 发际线→下巴尖（不超出）；v1 = 头顶→下巴尖并两端超出一点
-    v_top = fy0 if variant == "v6" else fy0 - over
-    v_bot = fy1 if variant == "v6" else fy1 + over

    # --- 1. 横向分界线（渐变，覆盖头宽并超出一点） ---
    for cy in ys:
@@ -7,7 +7,7 @@ Wants=comfyui.service change_hair-hair.service
 Type=simple
 User=ubuntu
 WorkingDirectory=/home/ubuntu/hair
-ExecStart=/home/ubuntu/hair/venv/bin/uvicorn app:app --host 0.0.0.0 --port 8187
+ExecStart=/home/ubuntu/hair/venv/bin/uvicorn app:app --host 0.0.0.0 --port 8187 --no-proxy-headers
 Restart=on-failure
 RestartSec=5

@@ -20,6 +20,13 @@ MP_TOP_ANCHORS: list[int] = [
 ]
 N_ANCHORS = len(MP_TOP_ANCHORS)  # 17

+# Anatomical left/right pairs, in MediaPipe indices. The anchor list is
+# NOT mirror-ordered at its ends: 127 pairs with 356, 234 with 454.
+MP_TOP_ANCHOR_PAIRS = [
+    (127, 356), (234, 454), (162, 389), (21, 251),
+    (54, 284), (103, 332), (67, 297), (109, 338),
+]
+
 # Vertex ID layout in the extended array of length 468 + 2*N_ANCHORS.
 #   [0       .. 468)         : MediaPipe canonical landmarks
 #   [468     .. 468+N)       : middle row (between MP boundary and hairline)
@@ -44,18 +44,18 @@ class FaceLandmarker:
        self._detector = vision.FaceLandmarker.create_from_options(options)
        self._mp = mp

-    def detect(self, image_rgb: np.ndarray) -> np.ndarray | None:
-        """Returns (468, 3) float32 of normalized x, y and relative z, or None.
+    def detect(self, image_rgb: np.ndarray, include_iris: bool = False) -> np.ndarray | None:
+        """Returns (468, 3), or (478, 3) with iris, normalized float32 points.

-        The task model produces 478 landmarks (468 face + 10 iris); we return
-        only the first 468 to match the canonical FaceMesh topology used by
-        the SDK's OBJ file.
+        The task model produces 478 landmarks (468 face + 10 iris). By default
+        only the first 468 match the SDK's canonical FaceMesh topology;
+        include_iris=True retains all 478 for face-measurement calibration.
        """
        mp_image = self._mp.Image(image_format=self._mp.ImageFormat.SRGB, data=image_rgb)
        result = self._detector.detect(mp_image)
        if not result.face_landmarks:
            return None
-        lm = result.face_landmarks[0][:468]
+        lm = result.face_landmarks[0] if include_iris else result.face_landmarks[0][:468]
        arr = np.array([[p.x, p.y, p.z] for p in lm], dtype=np.float32)
        return arr

@@ -0,0 +1,218 @@
+"""Interface 52: upright face axis, then native-resolution reference framing."""
+from __future__ import annotations
+
+from functools import lru_cache
+from io import BytesIO
+import json
+from pathlib import Path
+
+import cv2
+import numpy as np
+from PIL import Image, ImageOps
+
+PROFILE_PATH = Path(__file__).parent / "reference/framing_profile.json"
+
+
+def decode_portrait(raw):
+    """Respect phone EXIF orientation before detecting or transforming points."""
+    try:
+        with Image.open(BytesIO(raw)) as image:
+            return cv2.cvtColor(np.array(ImageOps.exif_transpose(image).convert("RGB")), cv2.COLOR_RGB2BGR)
+    except Exception:
+        return None
+
+
+@lru_cache(maxsize=1)
+def load_profile():
+    return json.loads(PROFILE_PATH.read_text(encoding="utf-8"))
+
+
+def transform_points(points, matrix):
+    return np.asarray(points, np.float64) @ matrix[:, :2].T + matrix[:, 2]
+
+
+def _rotation(landmarks, width, height):
+    xy = landmarks[:, :2].astype(np.float64) * [width, height]
+    delta = xy[152] - xy[9]
+    length = float(np.linalg.norm(delta))
+    if not np.isfinite(xy).all() or length < 2 or delta[1] <= 0:
+        raise ValueError("9号点与152号点无效，无法扶正人脸")
+    down = delta / length
+    rotation = np.array([[down[1], -down[0]], [down[0], down[1]]], np.float64)
+    return xy, rotation, length
+
+
+def _matrix(rotation, source_anchor, target_anchor):
+    return np.column_stack((rotation, np.asarray(target_anchor) - rotation @ source_anchor))
+
+
+def crop_fits(matrix, source_size, output_size):
+    w, h = output_size
+    corners = transform_points([[0, 0], [w-1, 0], [w-1, h-1], [0, h-1]], cv2.invertAffineTransform(matrix))
+    sw, sh = source_size
+    return bool((corners[:, 0] >= -1e-7).all() and (corners[:, 0] <= sw-1+1e-7).all()
+                and (corners[:, 1] >= -1e-7).all() and (corners[:, 1] <= sh-1+1e-7).all())
+
+
+def _inscribed_frame(rotation, source_size, aspect=None):
+    """Largest centred valid rectangle: no generated background or upscaling."""
+    sw, sh = source_size
+    ratio = sw / sh if aspect is None else aspect
+    source_center = np.array([(sw-1)/2, (sh-1)/2])
+    low, high = 2.0, min(float(sh), sw/ratio)
+    for _ in range(48):
+        h = (low+high)/2
+        w = h*ratio
+        matrix = _matrix(rotation, source_center, [(w-1)/2, (h-1)/2])
+        if crop_fits(matrix, source_size, (w, h)):
+            low = h
+        else:
+            high = h
+    if aspect == .75:
+        h = max(4, int(low+1e-7)//4*4)
+        w = h//4*3
+    else:
+        h, w = max(2, int(low+1e-7)), max(2, int(low*ratio+1e-7))
+    matrix = _matrix(rotation, source_center, [(w-1)/2, (h-1)/2])
+    if not crop_fits(matrix, source_size, (w, h)):
+        raise ValueError("图片边缘空间不足，无法得到有效扶正画面")
+    return matrix, (w, h)
+
+
+def _tall_portrait_frame(xy, rotation, source_size, profile):
+    """Largest native 3:4 frame, positioned to retain the entire face."""
+    _, size = _inscribed_frame(rotation, source_size, aspect=.75)
+    half = (np.array(size, np.float64)-1)/2
+    extent = np.abs(rotation.T) @ half
+    low, high = extent, np.array(source_size)-1-extent
+    polygon = [np.array(p) for p in ((low[0], low[1]), (high[0], low[1]),
+                                   (high[0], high[1]), (low[0], high[1]))]
+    face = xy[:468] @ rotation.T
+    # The crop centre must satisfy both the source boundaries and the face
+    # bounds. Clip the feasible polygon in double precision to avoid padding.
+    face_low, face_high = face.max(axis=0)-half, face.min(axis=0)+half
+    for axis in range(2):
+        for normal, bound in ((rotation[axis], face_high[axis]),
+                              (-rotation[axis], -face_low[axis])):
+            clipped = []
+            for start, end in zip(polygon, polygon[1:]+polygon[:1]):
+                a, b = float(normal @ start-bound-1e-8), float(normal @ end-bound-1e-8)
+                if (a <= 0) != (b <= 0):
+                    clipped.append(start+(end-start)*(a/(a-b)))
+                if b <= 0:
+                    clipped.append(end)
+            polygon = clipped
+            if not polygon:
+                raise ValueError("3:4裁切窗口无法保留完整人脸")
+    target = np.array([profile["brow_x_ratio"]*size[0], profile["brow_y_ratio"]*size[1]])
+    preferred = xy[9]-rotation.T @ (target-half)
+    # Euclidean projection keeps the brow as close as possible to its target.
+    candidates = []
+    if (preferred >= low-1e-8).all() and (preferred <= high+1e-8).all():
+        upright = rotation @ preferred
+        if (upright >= face_low-1e-8).all() and (upright <= face_high+1e-8).all():
+            candidates.append(preferred)
+    for start, end in zip(polygon, polygon[1:]+polygon[:1]):
+        edge = end-start
+        length2 = float(edge @ edge)
+        t = np.clip(float((preferred-start) @ edge)/length2, 0, 1) if length2 > 1e-16 else 0
+        candidates.append(start+t*edge)
+    centre = min(candidates, key=lambda p: float(np.sum((p-preferred)**2)))
+    matrix = _matrix(rotation, centre, half)
+    if not crop_fits(matrix, source_size, size):
+        raise ValueError("3:4裁切窗口超出原图有效范围")
+    return matrix, size
+
+
+def _apply(photo, landmarks, matrix, size):
+    h, w = photo.shape[:2]
+    ow, oh = size
+    moved = np.asarray(landmarks, np.float64).copy()
+    moved[:, :2] = transform_points(landmarks[:, :2]*[w, h], matrix) / [ow, oh]
+    # MediaPipe Z is measured in normalized image-width units.
+    moved[:, 2] *= w/ow
+    result = cv2.warpAffine(photo, matrix, size, flags=cv2.INTER_LINEAR,
+                            borderMode=cv2.BORDER_CONSTANT, borderValue=(0, 0, 0))
+    return result, moved
+
+
+def calibrate_reference(photo, landmarks):
+    """First straighten the actual reference; derive ratios from its final frame."""
+    h, w = photo.shape[:2]
+    _xy, rotation, length = _rotation(landmarks, w, h)
+    matrix, size = _inscribed_frame(rotation, (w, h), aspect=.75)
+    image, moved = _apply(photo, landmarks, matrix, size)
+    ow, oh = size
+    b, c = moved[[9, 152], :2]*[ow, oh]
+    profile = {
+        "version": "2026-11-04-upright-9-152-subject-90pct-tall-3x4",
+        "source_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/misc/20261103_user.jpg",
+        "reference_image_url": "/static/framing_reference.png",
+        "reference_size": {"width": ow, "height": oh},
+        "source_size": {"width": w, "height": h},
+        "brow_9": {"x": float(b[0]), "y": float(b[1])},
+        "chin_152": {"x": float(c[0]), "y": float(c[1])},
+        "face_axis_length_px": length,
+        "brow_x_ratio": float(b[0]/ow), "brow_y_ratio": float(b[1]/oh),
+        "reference_width_per_face_length": ow/length,
+        "reference_height_per_face_length": oh/length,
+        "width_per_face_length": 2.5, "height_per_face_length": 3.0/.9,
+        "rotation_degrees": float(np.degrees(np.arctan2(rotation[0, 1], rotation[0, 0]))),
+        "axis_x_error_px": float(abs(b[0]-c[0])),
+        "source_to_reference": matrix.tolist(),
+        "landmark_source": "source_landmarks_transformed_without_redetection",
+    }
+    return image, moved, profile
+
+
+def normalize_portrait(photo, landmarks, profile=None):
+    profile = load_profile() if profile is None else profile
+    h, w = photo.shape[:2]
+    xy, rotation, length = _rotation(landmarks, w, h)
+    # Round to an exact 3:4 frame without changing pixel scale.
+    target_h = max(4, round(profile["height_per_face_length"]*length/4)*4)
+    target_w = target_h//4*3
+    matrix = _matrix(rotation, xy[9], [profile["brow_x_ratio"]*target_w,
+                                      profile["brow_y_ratio"]*target_h])
+    cropped = crop_fits(matrix, (w, h), (target_w, target_h))
+    aspect_cropped = False
+    aspect_skip_reason = None
+    if cropped:
+        size = (target_w, target_h)
+    elif h*3 > w*4:
+        try:
+            matrix, size = _tall_portrait_frame(xy, rotation, (w, h), profile)
+            aspect_cropped = True
+        except ValueError as ex:
+            aspect_skip_reason = str(ex)
+            matrix, size = _inscribed_frame(rotation, (w, h))
+    else:
+        matrix, size = _inscribed_frame(rotation, (w, h))
+    image, moved = _apply(photo, landmarks, matrix, size)
+    ow, oh = size
+    axis = moved[[9, 152], :2]*[ow, oh]
+    if not ((axis[:, 0] >= 0).all() and (axis[:, 0] < ow).all()
+            and (axis[:, 1] >= 0).all() and (axis[:, 1] < oh).all()):
+        raise ValueError("人脸过于贴近边缘，扶正后无法保留9号点与152号点")
+    metadata = {
+        "enabled": True, "standard_crop": cropped, "aspect_crop": aspect_cropped,
+        "aspect_crop_skip_reason": aspect_skip_reason,
+        "status": "reference_crop" if cropped else ("aspect_crop" if aspect_cropped else "rotation_only"),
+        "skip_reason": None if cropped else "insufficient_space",
+        "message": ("已扶正并按参考图裁切" if cropped else
+                    "原图偏长，已扶正并裁成3:4，保留完整人脸" if aspect_cropped else
+                    "3:4窗口无法完整保留人脸，已仅扶正" if aspect_skip_reason else
+                    "空间不足，仅扶正，未按参考比例裁切"),
+        "rotation_degrees": float(np.degrees(np.arctan2(rotation[0, 1], rotation[0, 0]))),
+        "source_size": {"width": w, "height": h}, "output_size": {"width": ow, "height": oh},
+        "requested_crop_size": {"width": target_w, "height": target_h},
+        "face_axis_length_px": length,
+        "height_per_face_length": profile["height_per_face_length"],
+        "width_per_face_length": profile["height_per_face_length"]*.75,
+        "source_to_processed": matrix.tolist(),
+        "processed_to_source": cv2.invertAffineTransform(matrix).tolist(),
+        "axis_x_error_px": float(abs(axis[0, 0]-axis[1, 0])),
+        "profile_version": profile["version"],
+        "reference_image_url": profile["reference_image_url"],
+    }
+    return image, moved, metadata
@@ -124,6 +124,46 @@ def build_hairline_mask(parse_map: np.ndarray) -> np.ndarray:
    return build_hair_mask(parse_map)


+def align_hairline_for_template(
+    landmarks_norm: np.ndarray,
+    hairline_norm: np.ndarray,
+    valid: np.ndarray,
+    image_shape: tuple[int, int],
+) -> np.ndarray:
+    """Keep a designed template aligned with the face rather than the fringe.
+
+    Uneven bangs can make the detected hair boundary much higher on one
+    side, shearing an otherwise balanced template. Average the upward
+    extension of each anatomical anchor pair, measured in image pixels.
+    Keep the face anchors and the perpendicular component unchanged so
+    facial roll/perspective remain visible; do not level the curve in the
+    image. Missing ray hits retain their fallback and never influence a
+    valid partner. Call silhouette clamping afterwards to bound the result.
+
+    This is for template placement only, not measurement of the real hairline.
+    """
+    h, w = image_shape
+    scale = np.array([w, h], dtype=np.float64)
+    up = face_up_vector(landmarks_norm) * scale
+    length = np.linalg.norm(up)
+    if length < 1e-9:
+        return hairline_norm.copy()
+    up /= length
+
+    anchors = landmarks_norm[C.MP_TOP_ANCHORS, :2] * scale
+    points = hairline_norm * scale
+    distances = (points - anchors) @ up
+    indices = {mp_idx: i for i, mp_idx in enumerate(C.MP_TOP_ANCHORS)}
+    for left, right in C.MP_TOP_ANCHOR_PAIRS:
+        i, j = indices[left], indices[right]
+        if not (valid[i] and valid[j]):
+            continue
+        mean_distance = (distances[i] + distances[j]) / 2.0
+        for k in (i, j):
+            points[k] += (mean_distance - distances[k]) * up
+    return (points / scale).astype(np.float32)
+
+
 def smooth_hairline(
    hairline_norm: np.ndarray,
    valid: np.ndarray,
@@ -0,0 +1,117 @@
+"""接口52设计高度：上庭 = (中庭 + 下庭) / 2；不改变接口2/5的几何。"""
+from __future__ import annotations
+
+import cv2
+import numpy as np
+
+from .mask import mask_from_curve
+
+
+def height_target(landmarks, width, height, reference=None):
+    """Use unrounded measurement coordinates; fall back to the same MP indices.
+
+    Distances use image Y, matching face_analysis.measure's court heights.
+    No rounded centimetre values or existing-hairline estimates enter the rule.
+    """
+    scale = np.array([width, height], dtype=np.float64)
+    if reference is None:
+        brow, nose, chin = landmarks[[9, 94, 152], :2] * scale
+        source = "hairline_landmarks"
+        px_per_cm = None
+    else:
+        brow = np.asarray(reference["brow_center"], dtype=np.float64)
+        nose = np.asarray(reference["nose_bottom"], dtype=np.float64)
+        chin = np.asarray(reference["chin_tip"], dtype=np.float64)
+        source = "face_measure"
+        px_per_cm = reference.get("px_per_cm")
+    if not np.isfinite([brow, nose, chin]).all() or not brow[1] < nose[1] < chin[1]:
+        raise ValueError("眉心、鼻翼下缘与下巴坐标异常，无法计算目标上庭高度")
+    middle, lower = float(nose[1] - brow[1]), float(chin[1] - nose[1])
+    upper = (middle + lower) / 2.0
+    target_y = float(brow[1] - upper)
+    if not (0 <= brow[0] < width and 1 <= target_y < height - 1):
+        raise ValueError("按中下庭均值计算的目标发际线超出图片范围，请使用完整正面照")
+    def pt(p):
+        return {"x": float(p[0]), "y": float(p[1])}
+    return {
+        "name": "court_average",
+        "formula": "上庭 = (中庭 + 下庭) / 2",
+        "measurement_source": source,
+        "brow_center": pt(brow), "nose_bottom": pt(nose), "chin_tip": pt(chin),
+        "middle_height_px": middle, "lower_height_px": lower,
+        "target_upper_height_px": upper,
+        "target_upper_height_cm": upper / px_per_cm if px_per_cm else None,
+        "target_center": {"x": float(brow[0]), "y": target_y},
+    }
+
+
+def clean_curve(overlay):
+    """Keep the white design line; discard opaque black texture-border pixels."""
+    out = np.zeros_like(overlay, dtype=np.float32)
+    out[:, :, :3] = 255
+    out[:, :, 3] = overlay[:, :, 3] * (overlay[:, :, :3].mean(axis=2) > 128)
+    return out
+
+
+def curve_center(overlay, center_x):
+    """Alpha-weighted centre of the line at the facial midline, in image pixels."""
+    h, w = overlay.shape[:2]
+    column = int(round(center_x))
+    if not 0 <= column < w:
+        raise ValueError("面部中轴超出图片范围")
+    alpha = overlay[:, column, 3].astype(np.float64)
+    alpha[alpha <= 1] = 0
+    if not alpha.any():
+        # Rasterized endpoints may leave a 1px gap, but distant sideburns
+        # must never be silently substituted for a missing central curve.
+        columns = np.flatnonzero((overlay[:, :, 3] > 1).any(axis=0))
+        if not columns.size or np.min(np.abs(columns - column)) > 2:
+            raise ValueError("发际线模板在面部中轴处没有有效曲线")
+        column = int(columns[np.argmin(np.abs(columns - column))])
+        alpha = overlay[:, column, 3].astype(np.float64)
+        alpha[alpha <= 1] = 0
+    return {"x": float(center_x), "y": float(np.dot(np.arange(h), alpha) / alpha.sum())}
+
+
+def place_curve(overlay, target, landmarks):
+    """Rigidly shift a rendered template along face-up to meet the target Y.
+
+    Keeping the curve rigid preserves roll and the template's shape. Pixel Y
+    remains the height convention, even for a rolled face. A short correction
+    compensates for the curve's slope as the local-axis shift also changes X.
+    """
+    curve = clean_curve(overlay)
+    h, w = curve.shape[:2]
+    center_x, target_y = target["x"], target["y"]
+    before = curve_center(curve, center_x)
+    up = (landmarks[10, :2] - landmarks[152, :2]) * [w, h]
+    if abs(up[1]) < 1e-6:
+        raise ValueError("无法确定人脸向上方向")
+    x_per_y = float(np.clip(up[0] / up[1], -1.0, 1.0))
+    dy = target_y - before["y"]
+    moved = curve
+    for _ in range(8):
+        transform = np.array([[1, 0, dy * x_per_y], [0, 1, dy]], np.float32)
+        moved = cv2.warpAffine(curve, transform, (w, h), flags=cv2.INTER_LINEAR,
+                               borderMode=cv2.BORDER_CONSTANT, borderValue=0)
+        center = curve_center(moved, center_x)
+        residual = target_y - center["y"]
+        if abs(residual) <= 0.15:
+            break
+        dy += residual
+    if abs(center["y"] - target_y) > 1.0:
+        raise ValueError("发际线中心无法对齐目标高度，请检查照片姿态与裁剪范围")
+    return moved, {
+        "old_center": before, "new_center": center,
+        "shift_x_px": float(dy * x_per_y), "shift_y_px": float(dy),
+        "target_error_px": float(abs(center["y"] - target_y)),
+    }
+
+
+def inpaint_from_curve(photo, landmarks, parse_map, overlay):
+    """The preview line is also the exact boundary used for the black line/mask."""
+    alpha = np.clip(overlay[:, :, 3:4] / 255.0, 0, 1)
+    marked = np.clip(photo * (1.0 - alpha), 0, 255).astype(np.uint8)
+    curve = (overlay[:, :, 3] > 40).astype(np.uint8) * 255
+    mask = mask_from_curve(curve, landmarks, parse_map)
+    return marked, mask
@@ -0,0 +1,43 @@
+{
+  "version": "2026-11-04-upright-9-152-subject-90pct-tall-3x4",
+  "source_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/misc/20261103_user.jpg",
+  "reference_image_url": "/static/framing_reference.png",
+  "reference_size": {
+    "width": 786,
+    "height": 1048
+  },
+  "source_size": {
+    "width": 792,
+    "height": 1056
+  },
+  "brow_9": {
+    "x": 385.1878369059212,
+    "y": 360.65437126595765
+  },
+  "chin_152": {
+    "x": 385.18783690592113,
+    "y": 725.8251047982238
+  },
+  "face_axis_length_px": 365.1707335322661,
+  "brow_x_ratio": 0.4900608612034621,
+  "brow_y_ratio": 0.3441358504446161,
+  "reference_width_per_face_length": 2.152417835890317,
+  "reference_height_per_face_length": 2.8698904478537566,
+  "width_per_face_length": 2.5,
+  "height_per_face_length": 3.333333333333333,
+  "rotation_degrees": 0.2596248408989324,
+  "axis_x_error_px": 5.684341886080802e-14,
+  "source_to_reference": [
+    [
+      0.9999897336401403,
+      0.004531292786961665,
+      -5.386196599797756
+    ],
+    [
+      -0.004531292786961665,
+      0.9999897336401403,
+      -2.202458197930582
+    ]
+  ],
+  "landmark_source": "source_landmarks_transformed_without_redetection"
+}
@@ -15,12 +15,14 @@ from . import comfyui
 from .face_landmarks import FaceLandmarker
 from .face_parsing import FaceParser
 from .hairline_2d import (
-    smooth_hairline, sample_hairline_clamped,
+    smooth_hairline, sample_hairline, align_hairline_for_template,
+    clamp_hairline_to_silhouette,
 )
 from .lift_3d import lift_hairline_to_3d, build_middle_row, assemble_full
 from .render import load_ext_mesh, load_texture_rgba, render_hairline_overlay, build_overlay_layer
 from .mask import build_inpaint_mask, compose_comfy_rgba, mask_from_curve
 from .marker_detect import detect_marker_hairline, path_to_curve_mask
+from .height_rule import height_target, clean_curve, place_curve, inpaint_from_curve

 import base64
 import io
@@ -177,20 +179,25 @@ def extract_502(image_bgr: np.ndarray):
    return ctx["points"], ctx["valid"]


-def extract_context(image_bgr: np.ndarray):
+def extract_context(image_bgr: np.ndarray, landmarks: np.ndarray | None = None):
    """照片(BGR) → {landmarks, parse_map, points, valid}。无人脸返回 None。

-    发际线几何检测固定用 `sample_hairline_clamped`（射线检测 + 头部轮廓钳制）：
+    发际线几何使用射线检测 + 人脸对应点对齐 + 头部轮廓钳制：
+    左右刘海高低不同时，协调对应锚点的延伸距离，避免把设计模板拉歪；
+    保留人脸本身的倾斜。预览白线与生发黑线/遮罩共用同一套 points。
    短发/剃光头照片（如 man_test.jpg）中间锚点检测失效时，纯射线检测的固定 fallback
    偏移会把点顶到头部轮廓外面的背景，产生"发际线贴到头部外面"的视觉 bug；钳制兜底后
-    fallback 点不会再跑出头部轮廓，正常长发照片结果与旧行为一致。
+    fallback 点不会再跑出头部轮廓。
    """
    rgb = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)
-    landmarks = get_landmarker().detect(rgb)
+    if landmarks is None:
+        landmarks = get_landmarker().detect(rgb)
    if landmarks is None:
        return None
    parse_map = get_parser().parse(rgb)
-    hairline_2d, valid = sample_hairline_clamped(landmarks, parse_map)
+    hairline_2d, valid = sample_hairline(landmarks, parse_map)
+    hairline_2d = align_hairline_for_template(landmarks, hairline_2d, valid, parse_map.shape)
+    hairline_2d = clamp_hairline_to_silhouette(hairline_2d, parse_map)
    hairline_2d = smooth_hairline(hairline_2d, valid)
    hairline_3d = lift_hairline_to_3d(landmarks, hairline_2d)
    middle_3d = build_middle_row(landmarks, hairline_3d)
@@ -533,6 +540,79 @@ def generate_hairline_pngs(image_bgr: np.ndarray, gender: str,
    return {"images": images, "best_centers": best_centers}


+def generate_hairline_pngs_52(image_bgr: np.ndarray, gender: str,
+                           hair_styles: list[int], use_mask: bool = True,
+                           prompt: str | None = None,
+                           generate_grow_image: bool = True,
+                           height_reference: dict | None = None,
+                           compare_rules: bool = False,
+                           precomputed_landmarks: np.ndarray | None = None):
+    """接口52：每种 middle 曲线中点对齐到眉心上方 (中庭+下庭)/2。
+
+    只读取 middle 贴图，high/low 保留空值。生发直接复用移动后的预览曲线，
+    不再使用位置不同的黑模板。compare_rules 仅为调试页附带旧规则叠图。
+    接口2仍使用原有管线。
+    """
+    if gender not in ("male", "female"):
+        raise ValueError(f"gender 必须是 male/female，收到 {gender!r}")
+    if not hair_styles:
+        raise ValueError("hair_styles 必填且不能为空")
+    ctx = extract_context(image_bgr, landmarks=precomputed_landmarks)
+    if ctx is None:
+        return None
+    h, w = image_bgr.shape[:2]
+    uv, ext_faces = load_ext_mesh()
+    rule = height_target(ctx["landmarks"], w, h, height_reference)
+    textures = get_texture_map()[gender]
+
+    # use_mask=False：干净原图+空遮罩与贴图无关，只跑一次 ComfyUI，选中项复用
+    # generate_grow_image=False：完全跳过生发（最耗时），grown_png 恒为 None
+    shared_grown = None
+    if generate_grow_image and not use_mask:
+        shared_grown = _grow_from_texture(image_bgr, ctx, None, use_mask=False, prompt=prompt)
+
+    images, best_centers = [], None
+    for s in hair_styles:  # s = 1-indexed 发型序号
+        key, mid_path = textures[s - 1]
+        old = build_overlay_layer(h, w, ctx["points"], ext_faces, uv, load_texture_rgba(mid_path))
+        overlay, placement = place_curve(old, rule["target_center"], ctx["landmarks"])
+        overlays = {"middle": overlay, "high": None, "low": None}
+        if not generate_grow_image:
+            grown_png = None
+        elif not use_mask:
+            grown_png = shared_grown
+        else:
+            grown_png = None
+            try:
+                marked, mask = inpaint_from_curve(
+                    image_bgr, ctx["landmarks"], ctx["parse_map"], overlay)
+                buf = io.BytesIO()
+                compose_comfy_rgba(marked, mask).save(buf, format="PNG", compress_level=1)
+                grown_png = comfyui.run(buf.getvalue(), prompt=prompt)
+            except Exception as ex:  # 单张生发失败不丢弃设计曲线
+                logger.warning("接口52 生发失败 type=%s：%s", key, ex)
+        # Flux may round dimensions to a multiple of its latent stride.
+        # Restore the input size so the preview and generated photo align.
+        if grown_png:
+            grown_png = _upscale_png_to(grown_png, w, h)
+        item = {"hairline_type": key, "order": s, "overlays": overlays,
+                "grown_png": grown_png, "height_placement": placement}
+        if compare_rules:
+            item["legacy_overlay"] = clean_curve(old)
+        images.append(item)
+        if best_centers is None:
+            center = placement["new_center"]
+            best_centers = {"middle": (round(center["x"]), round(center["y"])),
+                            "high": None, "low": None}
+    guide_landmarks = {
+        str(index): {"x": float(ctx["landmarks"][index, 0] * w),
+                     "y": float(ctx["landmarks"][index, 1] * h)}
+        for index in (9, 150, 152, 162, 251, 389)
+    }
+    return {"images": images, "best_centers": best_centers, "height_rule": rule,
+            "guide_landmarks": guide_landmarks}
+
+
 # 接口3 送 ComfyUI 前限边，降低峰值显存，避免与接口2 切换时把 Flux 挤出。
 # 统一 prompt 后 Flux 不再被 CLIP 挤出，接口3 可用较高分辨率。可用 GROW_B_MAX_SIDE 覆盖。
 _GROW_B_MAX_SIDE = int(os.getenv("GROW_B_MAX_SIDE", "1024"))
@@ -0,0 +1,3 @@
+{
+  "allowed_ips": ["49.232.51.171"]
+}
@@ -0,0 +1,77 @@
+"""Direct-peer IP access policy for the GPU worker's dynamic endpoints."""
+import ipaddress
+import json
+from pathlib import Path
+import socket
+
+CONFIG_PATH = Path(__file__).resolve().parent / "ip_access.json"
+
+
+def _address(value):
+    address = ipaddress.ip_address(value)
+    return address.ipv4_mapped if isinstance(address, ipaddress.IPv6Address) and address.ipv4_mapped else address
+
+
+def local_addresses():
+    addresses = {_address("127.0.0.1"), _address("::1")}
+    try:
+        addresses.update(_address(item[4][0].split("%", 1)[0])
+                         for item in socket.getaddrinfo(socket.gethostname(), None))
+    except (OSError, ValueError):
+        pass
+    # Include all assigned Linux interface addresses, not entire private ranges.
+    try:
+        import fcntl
+        import struct
+        with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as sock:
+            for _, name in socket.if_nameindex():
+                try:
+                    data = fcntl.ioctl(sock.fileno(), 0x8915, struct.pack("256s", name.encode()[:15]))
+                    addresses.add(_address(socket.inet_ntoa(data[20:24])))
+                except OSError:
+                    continue
+        path = Path("/proc/net/if_inet6")
+        if path.exists():
+            addresses.update(ipaddress.IPv6Address(line.split()[0]) for line in path.read_text().splitlines())
+    except (ImportError, OSError, ValueError):
+        pass
+    return addresses
+
+
+def _networks(values):
+    if not isinstance(values, list) or not all(isinstance(value, str) for value in values):
+        raise ValueError("IP配置必须是字符串列表")
+    return [ipaddress.ip_network(value, strict=False) for value in values]
+
+
+def _matches(address, networks):
+    return any(address.version == network.version and address in network for network in networks)
+
+
+def check_access(request):
+    """Return (allowed, reason), re-reading the whitelist for each request."""
+    try:
+        config = json.loads(CONFIG_PATH.read_text()) if CONFIG_PATH.exists() else {}
+        allowed = _networks(config.get("allowed_ips", []))
+        if request.client is None or request.client.port == 0:
+            # Uvicorn proxy middleware has discarded the socket peer at port 0.
+            # Reject rather than treating an unverified forwarded IP as local.
+            return False, "无法确认直连来源，请使用 --no-proxy-headers 启动服务"
+        peer = _address(request.client.host)
+        # Authorize the gateway's socket IP, never its end-user forwarding headers.
+        if peer.is_unspecified:
+            return False, "来源地址无效"
+        return (True, "") if peer.is_loopback or peer in local_addresses() or _matches(peer, allowed) else (False, "来源IP不在接口白名单")
+    except (OSError, ValueError, TypeError, AttributeError):
+        return False, "IP白名单配置或来源地址无效"
+
+
+def rejection(request):
+    from fastapi.responses import JSONResponse
+    if request.url.path == "/static" or request.url.path.startswith("/static/"):
+        return None
+    allowed, reason = check_access(request)
+    if allowed:
+        return None
+    return JSONResponse(status_code=403, content={"code": 1010, "message": reason,
+                                               "request_id": "ip-access", "data": None})
@@ -31,11 +31,11 @@ fi

 if [ "${FOREGROUND:-0}" = "1" ]; then
  echo ">> 前台运行 worker: ${HOST}:${PORT}"
-  exec "$VENV_UVICORN" app:app --host "$HOST" --port "$PORT"
+  exec "$VENV_UVICORN" app:app --host "$HOST" --port "$PORT" --no-proxy-headers
 fi

 echo ">> 后台启动 worker: ${HOST}:${PORT}"
-nohup "$VENV_UVICORN" app:app --host "$HOST" --port "$PORT" >> "$LOG_FILE" 2>&1 &
+nohup "$VENV_UVICORN" app:app --host "$HOST" --port "$PORT" --no-proxy-headers >> "$LOG_FILE" 2>&1 &
 echo $! > "$PID_FILE"
 sleep 1
 if kill -0 "$(cat "$PID_FILE")" 2>/dev/null; then
@@ -0,0 +1,25 @@
+#!/usr/bin/env python3
+"""Straighten the specified reference first, then recalibrate framing ratios."""
+import json
+from pathlib import Path
+import sys
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+import cv2
+from hairline.framing import calibrate_reference, decode_portrait, PROFILE_PATH
+from hairline.service import get_landmarker
+
+if __name__ == "__main__":
+    source = Path(sys.argv[1]) if len(sys.argv)>1 else ROOT/"tests/output/upload_framing/reference.jpg"
+    photo = decode_portrait(source.read_bytes())
+    if photo is None:
+        raise ValueError("Reference image cannot be decoded")
+    landmarks = get_landmarker().detect(cv2.cvtColor(photo, cv2.COLOR_BGR2RGB), include_iris=True)
+    if landmarks is None:
+        raise ValueError("Reference face not detected")
+    image, _, profile = calibrate_reference(photo, landmarks)
+    cv2.imwrite(str(ROOT/"static/framing_reference.png"), image)
+    PROFILE_PATH.write_text(json.dumps(profile, ensure_ascii=False, indent=2)+"\n", encoding="utf-8")
+    print(json.dumps(profile, ensure_ascii=False, indent=2))
@@ -0,0 +1,122 @@
+#!/usr/bin/env python3
+"""保存接口52旧/新高度规则对比，供 static/test_interface52_compare.html 展示。
+
+默认验证用户清单全部30张，男女各7型，共420组；只跑定位，不调用ComfyUI。
+用法：./venv/bin/python scripts/compare_hairline_height.py
+可用 --cases 1,2,3,4,5 只验证指定案例。
+"""
+from __future__ import annotations
+
+import argparse
+import base64
+from datetime import datetime
+import json
+import os
+from pathlib import Path
+import time
+from zoneinfo import ZoneInfo
+
+import cv2
+import httpx
+import numpy as np
+
+ROOT = Path(__file__).resolve().parent.parent
+
+
+def run(args):
+    manifest = json.loads((ROOT / "tests/fixtures/hairline_height_cases.json").read_text())
+    available = {c["image_number"]: c for c in manifest["cases"]}
+    requested = list(available) if args.cases == "all" else [int(n) for n in args.cases.split(",")]
+    cases = [available[n] for n in requested]
+    inputs = ROOT / "tests/output/hairline_height_compare/inputs"
+    inputs.mkdir(parents=True, exist_ok=True)
+    output = ROOT / "static/annotations/hairline_height_compare"
+    output.mkdir(parents=True, exist_ok=True)
+    report = {
+        "created_at": datetime.now(ZoneInfo("Asia/Shanghai")).isoformat(),
+        "case_numbers": requested, "groups": [], "errors": [],
+    }
+    started = time.monotonic()
+    with httpx.Client(timeout=180, follow_redirects=True) as client:
+        for case in cases:
+            number = case["image_number"]
+            source = inputs / f"case_{number}.png"
+            if not source.exists():
+                response = client.get(case["image_url"])
+                response.raise_for_status()
+                source.write_bytes(response.content)
+            photo = cv2.imdecode(np.frombuffer(source.read_bytes(), np.uint8), cv2.IMREAD_COLOR)
+            if photo is None:
+                raise ValueError(f"案例 {number} 无法解码")
+            # Keep the source photo available even when positioning fails.
+            original_path = output / f"case_{number}_original.jpg"
+            cv2.imwrite(str(original_path), photo, [cv2.IMWRITE_JPEG_QUALITY, 95])
+            original_url = "/static/annotations/hairline_height_compare/" + original_path.name
+            # Interface 52 frames native-resolution input before any size changes.
+            content = cv2.imencode(".png", photo)[1].tobytes()
+            for gender in args.genders.split(","):
+                step = time.monotonic()
+                try:
+                    response = client.post(args.base_url.rstrip("/") + "/api/v1/debug/hairline-height-compare",
+                        files={"image_file": (f"case_{number}.png", content, "image/png")},
+                        data={"gender": gender, "hair_style": "1,2,3,4,5,6,7"})
+                    response.raise_for_status()
+                    body = response.json()
+                    if body["code"] != 0:
+                        raise ValueError(body["message"])
+                    group = body["data"]
+                    group.update(case_number=number, source_url=case["image_url"],
+                                 elapsed_seconds=round(time.monotonic() - step, 2))
+                    def save_image(obj, field, suffix):
+                        value = obj.pop(field + "_base64", None)
+                        if not value:
+                            return
+                        # Both gender sets use the same source and processed
+                        # photos; share their URLs and browser image cache.
+                        prefix = f"case_{number}" if obj is group else f"case_{number}_{gender}"
+                        path = output / f"{prefix}_{suffix}"
+                        path.write_bytes(base64.b64decode(value.split(",", 1)[-1]))
+                        obj[field + "_url"] = "/static/annotations/hairline_height_compare/" + path.name
+                    save_image(group, "input", "input.jpg")
+                    save_image(group, "original", "original.jpg")
+                    for result in group["results"]:
+                        save_image(result, "old_overlay", f"{result['order']}_old.png")
+                        save_image(result, "new_overlay", f"{result['order']}_new.png")
+                        save_image(result, "grown_image", f"{result['order']}_grown.jpg")
+                        if result["height_placement"]["target_error_px"] > 1:
+                            raise ValueError(f"{result['hairline_type']} 中心误差超过1像素")
+                    if group["framing"]["axis_x_error_px"] > 1:
+                        raise ValueError("扶正后的9/152横向误差超过1像素")
+                    report["groups"].append(group)
+                    print(f"case {number} / {gender}: {len(group['results'])} shapes OK ({group['elapsed_seconds']}s)", flush=True)
+                except Exception as ex:
+                    report["errors"].append({"case_number": number, "gender": gender, "message": str(ex),
+                                             "source_url": case["image_url"], "original_url": original_url,
+                                             "width": photo.shape[1], "height": photo.shape[0]})
+                    print(f"case {number} / {gender}: FAILED {ex}", flush=True)
+                # Atomically expose partial results to the page while running.
+                temporary = output / "latest.json.tmp"
+                temporary.write_text(json.dumps(report, ensure_ascii=False, indent=2))
+                os.replace(temporary, output / "latest.json")
+    results = [r for g in report["groups"] for r in g["results"]]
+    summary = {
+        "cases": len(cases), "successful_groups": len(report["groups"]),
+        "comparisons": len(results), "failures": len(report["errors"]),
+        "max_center_error_px": max((r["height_placement"]["target_error_px"] for r in results), default=None),
+        "elapsed_seconds": round(time.monotonic() - started, 2),
+    }
+    report["summary"] = summary
+    temporary = output / "latest.json.tmp"
+    temporary.write_text(json.dumps(report, ensure_ascii=False, indent=2))
+    os.replace(temporary, output / "latest.json")
+    print(json.dumps(summary, ensure_ascii=False), flush=True)
+    if report["errors"]:
+        raise SystemExit(1)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--cases", default="all", help="all 或逗号分隔的图片编号")
+    parser.add_argument("--genders", default="female,male")
+    parser.add_argument("--base-url", default="http://127.0.0.1:8187")
+    run(parser.parse_args())
@@ -0,0 +1,254 @@
+"use strict";
+const $ = id => document.getElementById(id);
+const styleNames = {
+  female: ["椭圆形", "花瓣形", "心形", "直线形", "波浪形", "大花瓣形", "古典花瓣形"],
+  male: ["椭圆形", "倒弧形", "M 型", "直线形", "心形", "柔和花瓣形", "波浪形"]
+};
+let dataset = {groups: [], errors: []}, mode = "side", drawVersion = 0, busy = false, cardObserver = null;
+const inputImages = new Map();
+const rowSelections = new Map();
+const esc = value => String(value).replace(/[&<>"']/g, c => ({"&":"&amp;","<":"&lt;",">":"&gt;",'"':"&quot;","'":"&#39;"}[c]));
+const fixed = (value, n = 2) => Number(value).toFixed(n);
+const imageSource = (object, field, mime = "image/png") => object[field + "_url"] || window.resolveImgSrc(null, object[field + "_base64"], mime);
+function status(text, error = false) { $("status").textContent = text; $("status").classList.toggle("error", error); }
+function loadImage(src) {
+  return new Promise((resolve, reject) => {
+    const im = new Image();
+    im.onload = () => resolve(im); im.onerror = () => reject(new Error("无法读取对比图片，请重新运行案例")); im.src = src;
+  });
+}
+function inputImage(group) {
+  const src = imageSource(group, "input", "image/jpeg");
+  if (!inputImages.has(src)) inputImages.set(src, loadImage(src));
+  return inputImages.get(src);
+}
+function originalCard(group,caseName){
+  const src=imageSource(group,"original","image/jpeg")||group.source_url||imageSource(group,"input","image/jpeg");
+  const size=group.framing?.source_size||{width:group.width,height:group.height};
+  const dimensions=size.width&&size.height?`width="${size.width}" height="${size.height}"`:"";
+  const sizeLabel=size.width&&size.height?`原尺寸 ${size.width}×${size.height}`:"完整上传图片";
+  const panel=document.createElement("article");panel.className="original-reference-card";
+  panel.innerHTML=`<div class="case-header"><div><span class="case-number original-reference-label">未裁切原图</span><h3>${esc(caseName)} · 上传原图</h3><p>处理前完整图片</p></div></div><div class="metric-line">${sizeLabel}</div><div class="image-pair overlay"><figure><figcaption>未扶正、未裁切</figcaption><a href="${esc(src)}" target="_blank" rel="noopener"><img src="${esc(src)}" ${dimensions} alt="${esc(caseName)}未裁切原图" loading="lazy" decoding="async"></a></figure></div><div class="case-bottom"><span>保留完整画面</span><a href="${esc(src)}" target="_blank" rel="noopener">查看原图 ↗</a></div>`;
+  return panel;
+}
+function failureCard(errors){
+  const panel=document.createElement("article");panel.className="failed-reference-card";
+  panel.innerHTML='<h3>定位失败，未生成对比图</h3><p>左侧保留完整上传原图。</p>';
+  for(const error of errors){
+    const message=document.createElement("p");
+    message.textContent=`${error.gender==="female"?"女性":"男性"}7型：${error.message}`;
+    panel.append(message);
+  }
+  return panel;
+}
+function refreshStyleOptions() {
+  const old = $("styleFilter").value || "all", gender = $("genderFilter").value;
+  $("styleFilter").replaceChildren();
+  for (let i = 0; i < 7; i++) {
+    const label = gender === "all" ? `${i+1}. 女：${styleNames.female[i]} / 男：${styleNames.male[i]}` : `${i+1}. ${styleNames[gender][i]}`;
+    $("styleFilter").add(new Option(label, String(i+1)));
+  }
+  $("styleFilter").add(new Option("全部 7 种形状", "all"));
+  $("styleFilter").value = old;
+}
+function refreshCases() {
+  const selected = $("caseFilter").value;
+  const cases = [...new Set([...dataset.groups,...dataset.errors].map(g => String(g.case_number)))].sort((a,b)=>a.localeCompare(b,"en",{numeric:true}));
+  $("caseFilter").replaceChildren(new Option(`全部 ${cases.length} 个案例`, "all"));
+  for (const key of cases) $("caseFilter").add(new Option(key.startsWith("custom") ? "自定义图片" : `案例 ${key}`, key));
+  if (cases.includes(selected)) $("caseFilter").value = selected;
+}
+function refreshStats() {
+  const results = dataset.groups.flatMap(g => g.results);
+  const errors = results.map(r => r.height_placement.target_error_px);
+  $("statCases").textContent = new Set([...dataset.groups,...dataset.errors].map(g => g.case_number)).size;
+  $("statComparisons").textContent = results.length;
+  $("statError").textContent = errors.length ? `${fixed(Math.max(...errors))} px` : "—";
+  $("statStatus").textContent = busy ? "运行中" : (dataset.errors.length ? "有异常" : (results.length ? "已完成" : "待运行"));
+  $("runTime").textContent = dataset.created_at ? `更新于 ${new Date(dataset.created_at).toLocaleString("zh-CN", {timeZone:"Asia/Shanghai",hour12:false})}` : "尚无保存结果";
+}
+function cropBox(group) {
+  if (!$("zoomFace").checked) return [0, 0, group.width, group.height];
+  if (group.crop_box) return group.crop_box;
+  const rule = group.height_rule, upper = rule.target_upper_height_px, center = rule.target_center;
+  const oldY = group.results.map(r => r.height_placement.old_center.y);
+  const left = Math.max(0, center.x - 1.7 * upper), top = Math.max(0, Math.min(center.y, ...oldY) - .55 * upper);
+  const right = Math.min(group.width, center.x + 1.7 * upper), bottom = Math.min(group.height, rule.brow_center.y + .75 * upper);
+  return [left, top, right, bottom];
+}
+function paint(canvas, photo, oldLine, newLine, group, kind) {
+  const [x0,y0,x1,y1] = cropBox(group), cw = x1-x0, ch = y1-y0;
+  canvas.width = kind === "overlay" ? 1000 : 650;
+  canvas.height = Math.round(canvas.width * ch / cw);
+  const ctx = canvas.getContext("2d"), W = canvas.width, H = canvas.height;
+  ctx.drawImage(photo, x0,y0,cw,ch,0,0,W,H);
+  const tint = (line, color) => {
+    const layer = document.createElement("canvas"); layer.width=W; layer.height=H;
+    const ink=layer.getContext("2d"); ink.drawImage(line,x0,y0,cw,ch,0,0,W,H);
+    ink.globalCompositeOperation="source-in";ink.fillStyle=color;ink.fillRect(0,0,W,H);ctx.drawImage(layer,0,0);
+  };
+  if (kind !== "new") tint(oldLine, "#f3a338");
+  if (kind !== "old") tint(newLine, "#10c6b6");
+  if ($("showTarget").checked) {
+    const target = group.height_rule.target_center, tx=(target.x-x0)/cw*W, ty=(target.y-y0)/ch*H;
+    ctx.save();ctx.strokeStyle="rgba(255,255,255,.85)";ctx.lineWidth=1.7;ctx.setLineDash([7,6]);
+    ctx.beginPath();ctx.moveTo(0,ty);ctx.lineTo(W,ty);ctx.stroke();ctx.setLineDash([]);
+    ctx.fillStyle="#fff";ctx.beginPath();ctx.arc(tx,ty,4,0,2*Math.PI);ctx.fill();ctx.restore();
+  }
+  if(kind!=="old"&&$("showBox").checked&&group.guide_landmarks){
+    window.drawHairlineReferenceBox(ctx,group.guide_landmarks,[x0,y0,x1,y1],W,H);
+  }
+}
+async function render() {
+  if(cardObserver)cardObserver.disconnect();
+  const version=++drawVersion,gender=$("genderFilter").value,number=$("caseFilter").value,style=$("styleFilter").value;
+  const cases=new Map();
+  for(const group of dataset.groups){
+    if(gender!=="all"&&group.gender!==gender)continue;
+    if(number!=="all"&&String(group.case_number)!==number)continue;
+    for(const result of group.results){
+      if(style!=="all"&&result.order!==Number(style))continue;
+      if(!cases.has(group.case_number))cases.set(group.case_number,[]);
+      cases.get(group.case_number).push({group,result});
+    }
+  }
+  const visibleErrors=dataset.errors.filter(error=>(gender==="all"||error.gender===gender)&&(number==="all"||String(error.case_number)===number));
+  for(const error of visibleErrors)if(!cases.has(error.case_number))cases.set(error.case_number,[]);
+  $("results").replaceChildren();
+  let drawErrors=0,total=0;
+  const rowUpdates=[],drawJobs=new WeakMap();
+  cardObserver=new IntersectionObserver(entries=>{
+    for(const entry of entries)if(entry.isIntersecting){
+      cardObserver.unobserve(entry.target);drawJobs.get(entry.target)?.();
+    }
+  },{rootMargin:"600px 0px"});
+  for(const [caseNumber,cards] of [...cases].sort((a,b)=>String(a[0]).localeCompare(String(b[0]),"en",{numeric:true}))){
+    total+=cards.length;
+    const row=document.createElement("section");row.className="person-row";
+    const caseName=String(caseNumber).startsWith("custom")?"自定义图片":`案例 ${caseNumber}`;
+    const caseErrors=visibleErrors.filter(error=>error.case_number===caseNumber);
+    if(!cards.length){
+      row.classList.add("failed-case");
+      row.innerHTML=`<div class="person-row-header"><h2 class="case-group-heading">${esc(caseName)} · 定位失败</h2></div><div class="person-body"><div class="person-styles" role="region" tabindex="0" aria-label="${esc(caseName)}的原图与定位结果"></div></div>`;
+      row.querySelector(".person-styles").append(originalCard(caseErrors[0],caseName),failureCard(caseErrors));
+      $("results").append(row);continue;
+    }
+    row.innerHTML=`<div class="person-row-header"><h2 class="case-group-heading">${esc(caseName)} · ${cards.length} 种发际线样式</h2><div class="row-controls"><span>点击样式，左侧旧图同步切换</span><button type="button" aria-label="查看前面的样式">←</button><button type="button" aria-label="查看后面的样式">→</button></div></div><div class="person-body"><article class="old-reference-card"><div class="case-header"><div><span class="case-number old-reference-label"><span class="dot old"></span>旧规则 · 头发边界定位</span><h3 class="old-style-name">正在加载…</h3><p>与右侧选中的形状对应</p></div></div><div class="metric-line old-style-metrics"></div><div class="image-pair overlay"><figure><figcaption>当前选中样式的原位置</figcaption><canvas aria-label="旧规则参考图"></canvas></figure></div><div class="case-bottom">本行仅显示一次 · 点击右侧样式切换</div></article><div class="person-styles" role="region" tabindex="0" aria-label="${esc(caseName)}的全部新规则样式"></div></div>`;
+    $("results").append(row);
+    const groupForRow=cards[0].group,framing=groupForRow.framing;
+    if(framing){
+      const info=document.createElement("div");info.className="framing-info";
+      info.innerHTML=`${esc(framing.message)} · 旋转 ${fixed(framing.rotation_degrees,3)}° · ${framing.source_size.width}×${framing.source_size.height} → ${framing.output_size.width}×${framing.output_size.height} · 9/152 横向差 ${fixed(framing.axis_x_error_px,3)} px <a target="_blank" rel="noopener" href="${esc(imageSource(groupForRow,"original","image/jpeg")||groupForRow.source_url)}">上传原图 ↗</a> <a target="_blank" rel="noopener" href="${esc(imageSource(groupForRow,"input","image/jpeg"))}">处理后底图 ↗</a>`;
+      row.insertBefore(info,row.querySelector(".person-body"));
+    }
+    const strip=row.querySelector(".person-styles"),oldPanel=row.querySelector(".old-reference-card"),oldCanvas=oldPanel.querySelector("canvas");
+    // One horizontal row keeps the unprocessed photo first on narrow screens
+    // too, without squeezing the new styles into a zero-width column.
+    strip.append(originalCard(groupForRow,caseName),oldPanel);
+    if(caseErrors.length)strip.append(failureCard(caseErrors));
+    const [previous,next]=row.querySelectorAll(".row-controls button");
+    previous.addEventListener("click",()=>strip.scrollBy({left:-strip.clientWidth*.85,behavior:"smooth"}));
+    next.addEventListener("click",()=>strip.scrollBy({left:strip.clientWidth*.85,behavior:"smooth"}));
+    const updateArrows=()=>{previous.disabled=strip.scrollLeft<=2;next.disabled=strip.scrollWidth-strip.clientWidth-strip.scrollLeft<=2;};
+    strip.addEventListener("scroll",updateArrows,{passive:true});rowUpdates.push(updateArrows);
+    let selectionVersion=0;
+    const selectOld=async(group,result,drawNow=true)=>{
+      const selected=++selectionVersion,key=`${group.gender}:${result.order}`;
+      oldPanel.dataset.drawn="false";
+      rowSelections.set(caseNumber,key);row.dataset.selectedStyle=key;
+      for(const card of strip.querySelectorAll(".case-card")){
+        const active=card.dataset.styleKey===key;card.classList.toggle("selected",active);
+        card.setAttribute("aria-current",String(active));card.querySelector(".select-style").textContent=active?"已对照":"对照此样式";
+      }
+      const label=styleNames[group.gender][result.order-1];
+      oldPanel.querySelector(".old-style-name").textContent=`${group.gender==="female"?"女性":"男性"} · ${label}`;
+      oldPanel.querySelector(".old-style-metrics").textContent=`原中心 y ${fixed(result.height_placement.old_center.y)} · 目标 y ${fixed(group.height_rule.target_center.y)}`;
+      const box=cropBox(group);oldCanvas.width=650;oldCanvas.height=Math.round(650*(box[3]-box[1])/(box[2]-box[0]));
+      const draw=async()=>{
+        try{
+          const [photo,line]=await Promise.all([inputImage(group),loadImage(imageSource(result,"old_overlay"))]);
+          if(version!==drawVersion||selected!==selectionVersion)return;
+          paint(oldCanvas,photo,line,null,group,"old");oldPanel.dataset.drawn="true";
+        }catch(error){if(version===drawVersion)status(error.message,true);}
+      };
+      if(drawNow)await draw();
+      else{drawJobs.set(oldPanel,draw);cardObserver.observe(oldPanel);}
+    };
+    for(const {group,result} of cards){
+      const rule=group.height_rule,p=result.height_placement,label=styleNames[group.gender][result.order-1];
+      const card=document.createElement("article");card.className="case-card";card.tabIndex=0;
+      card.dataset.styleKey=`${group.gender}:${result.order}`;
+      card.innerHTML=`<div class="case-header"><div><span class="case-number">${group.gender==="female"?"女性":"男性"}发际线</span><h3>${esc(label)} <span style="font-size:11px;color:#7c8c88;font-weight:400">${esc(result.hairline_type)}</span></h3><p>目标上庭 (${fixed(rule.middle_height_px)} + ${fixed(rule.lower_height_px)}) ÷ 2 = ${fixed(rule.target_upper_height_px)} px</p></div></div><div class="metric-line"><span>新 y <b>${fixed(p.new_center.y)}</b></span><span><b>${p.shift_y_px<0?"上移":"下移"} ${fixed(Math.abs(p.shift_y_px))} px</b></span></div><div class="image-pair overlay"><figure><figcaption class="new"><span class="dot new"></span>${mode==="overlay"?"橙色旧线 / 青色新线":"新规则 · 中下庭均值"}</figcaption><canvas aria-label="新规则发际线"></canvas></figure></div><div class="case-bottom"><span>中心误差 ${fixed(p.target_error_px)} px</span><button class="select-style" type="button">对照此样式</button><a href="${esc(group.source_url||imageSource(group,"input","image/jpeg"))}" target="_blank" rel="noopener">原图 ↗</a></div>`;
+      strip.append(card);
+      card.addEventListener("click",event=>{if(!event.target.closest("a"))selectOld(group,result);});
+      card.addEventListener("keydown",event=>{if(event.target===card&&(event.key==="Enter"||event.key===" ")){event.preventDefault();selectOld(group,result);}});
+      const canvas=card.querySelector("canvas"),box=cropBox(group);canvas.width=650;canvas.height=Math.round(650*(box[3]-box[1])/(box[2]-box[0]));
+      drawJobs.set(card,async()=>{
+        try{
+          const [photo,line,oldLine]=await Promise.all([inputImage(group),loadImage(imageSource(result,"new_overlay")),mode==="overlay"?loadImage(imageSource(result,"old_overlay")):Promise.resolve(null)]);
+          if(version!==drawVersion)return;
+          paint(canvas,photo,oldLine,line,group,mode==="overlay"?"overlay":"new");card.dataset.drawn="true";
+        }catch(error){drawErrors++;const message=document.createElement("p");message.className="error-card";message.textContent=error.message;card.append(message);status(`有 ${drawErrors} 组图片加载失败，可重跑案例。`,true);}
+      });
+      cardObserver.observe(card);
+    }
+    const active=cards.find(c=>`${c.group.gender}:${c.result.order}`===rowSelections.get(caseNumber))||cards[0];
+    selectOld(active.group,active.result,false);
+  }
+  if(!cases.size&&!dataset.errors.length)$("results").innerHTML='<div class="empty">暂无对比结果。可上传图片或点击“重跑全部30个案例”。</div>';
+  requestAnimationFrame(()=>rowUpdates.forEach(update=>update()));
+  if(version===drawVersion&&!busy){
+    const failures=new Set(visibleErrors.map(error=>error.case_number)).size;
+    status(`显示 ${cases.size} 个案例、${total} 种新规则样式${failures?`，${failures} 个案例存在定位失败`:""}。每人一行，最前面为未裁切原图，旧图仅一次；左右滑动查看全部样式。`,failures>0);
+  }
+}
+function setBusy(value) {busy=value;$("rerunBatch").disabled=value;$("customRun").disabled=value;refreshStats();}
+async function compare(form) {
+  const response=await fetch("/api/v1/debug/hairline-height-compare",{method:"POST",body:form});
+  if(!response.ok)throw new Error(`请求失败：HTTP ${response.status}`);
+  const json=await response.json();if(json.code!==0)throw new Error(json.message);return json.data;
+}
+async function rerun() {
+  if(busy)return;setBusy(true);let failure=null;
+  try {
+    const response=await fetch("/api/v1/debug/hairline-height-cases");const json=await response.json();
+    if(json.code!==0)throw new Error(json.message);
+    dataset={groups:[],errors:[],created_at:new Date().toISOString()};inputImages.clear();
+    const cases=json.data.cases,total=cases.length*2;let done=0;
+    for(const test of cases)for(const gender of ["female","male"]){
+      status(`正在运行案例 ${test.image_number} / ${gender==="female"?"女性":"男性"} 7 型（${done}/${total}）…`);
+      const form=new FormData();form.append("image_url",test.image_url);form.append("gender",gender);form.append("hair_style","1,2,3,4,5,6,7");
+      try{const data=await compare(form);dataset.groups.push({...data,case_number:test.image_number,source_url:test.image_url});}
+      catch(error){dataset.errors.push({case_number:test.image_number,gender,message:error.message,source_url:test.image_url,original_url:test.image_url});}
+      done++;refreshCases();refreshStats();await render();
+    }
+  }catch(error){failure=error.message;}finally{setBusy(false);await render();if(failure)status(failure,true);}
+}
+async function custom(event) {
+  event.preventDefault();if(busy)return;const file=$("customFile").files[0];if(!file)return;setBusy(true);status("正在计算自定义图片的旧 / 新位置…");let failure=null;
+  try {
+    const photo=file,form=new FormData(),gender=$("customGender").value;
+    form.append("image_file",photo);form.append("gender",gender);form.append("hair_style","1,2,3,4,5,6,7");
+    const data=await compare(form),number=`custom-${Date.now()}`;
+    dataset.groups.push({...data,case_number:number});dataset.created_at=new Date().toISOString();
+    $("genderFilter").value=gender;refreshStyleOptions();refreshCases();$("caseFilter").value=number;
+  }catch(error){failure=error.message;}finally{setBusy(false);await render();if(failure)status(failure,true);}
+}
+async function init() {
+  fetch("/api/v1/debug/hairline-framing-reference").then(r=>r.json()).then(json=>{
+    if(json.code!==0)throw new Error(json.message);
+    const p=json.data;
+    $("framingStandard").innerHTML=`<img src="${esc(p.reference_image_url)}" alt="9号与152号点已竖直对齐的参考图" loading="lazy"><div><strong>已扶正参考图：${p.reference_size.width}×${p.reference_size.height}（3:4）</strong><p>参考图9号点 (${fixed(p.brow_9.x)}, ${fixed(p.brow_9.y)})<br>参考图152号点 (${fixed(p.chin_152.x)}, ${fixed(p.chin_152.y)})<br>两点X相同；旋转 ${fixed(p.rotation_degrees,3)}°</p><p><strong>当前裁切：高约 ${Number(fixed(p.height_per_face_length,4))}×L，宽 ${Number(fixed(p.width_per_face_length,4))}×L</strong>L 为9号点到152号点的像素距离。<br>眉心位置：横向 ${fixed(p.brow_x_ratio*100,2)}%，纵向 ${fixed(p.brow_y_ratio*100,2)}%</p><p>用户图先扶正，有空间时按上述尺寸裁切；长图保留人脸裁为3:4，其余空间不足时仅扶正。保留实际像素，不放大图片。</p></div>`;
+  }).catch(error=>{$("framingStandard").textContent=error.message;});
+  refreshStyleOptions();
+  $("genderFilter").addEventListener("change",()=>{refreshStyleOptions();render();});
+  for(const id of ["styleFilter","caseFilter","zoomFace","showTarget","showBox"])$(id).addEventListener("change",render);
+  for(const [id,value] of [["sideMode","side"],["overlayMode","overlay"]])$(id).addEventListener("click",()=>{mode=value;$("sideMode").classList.toggle("active",value==="side");$("overlayMode").classList.toggle("active",value==="overlay");render();});
+  $("customForm").addEventListener("submit",custom);$("rerunBatch").addEventListener("click",rerun);
+  try {
+    const response=await fetch("/static/annotations/hairline_height_compare/latest.json",{cache:"no-store"});
+    if(!response.ok)throw new Error("尚无保存的批量结果，可上传图片或重跑全部30个案例。");
+    dataset=await response.json();dataset.errors ||= [];refreshCases();refreshStats();await render();
+  }catch(error){refreshStats();status(error.message,true);}
+}
+document.addEventListener("DOMContentLoaded",init);
@@ -0,0 +1,62 @@
+"use strict";
+// Reference geometry is in input-image pixels; no screen-vertical assumptions.
+window.hairlineReferenceBox = function(points) {
+  const p = id => {
+    const value = points[String(id)];
+    if (!value || !Number.isFinite(value.x) || !Number.isFinite(value.y)) throw new Error(`缺少 ${id} 号关键点`);
+    return [value.x, value.y];
+  };
+  const add = (a,b) => [a[0]+b[0],a[1]+b[1]];
+  const sub = (a,b) => [a[0]-b[0],a[1]-b[1]];
+  const mul = (a,s) => [a[0]*s,a[1]*s];
+  const dot = (a,b) => a[0]*b[0]+a[1]*b[1];
+  const brow=p(9),chin=p(152),left=p(162),right=p(389);
+  const delta=sub(chin,brow),length=Math.hypot(...delta);
+  if(length<1)throw new Error("9→152 方向无法确定");
+  const up=mul(delta,-1/length),horizontal=[-up[1],up[0]],height=.7*length;
+  const sidePoint=(anchor,direction,level)=>{
+    const denominator=dot(direction,up);
+    if(Math.abs(denominator)<1e-6)throw new Error("参考框侧边与底边平行");
+    return add(anchor,mul(direction,(level-dot(sub(anchor,brow),up))/denominator));
+  };
+  const corners=[sidePoint(left,up,0),sidePoint(right,up,0),
+                 sidePoint(right,up,height),sidePoint(left,up,height)];
+  if(dot(sub(corners[1],corners[0]),horizontal)<=0)throw new Error("参考框左右位置异常");
+  return {corners,height,faceLength:length,
+          landmarks:{9:brow,152:chin,162:left,389:right}};
+};
+
+window.clipHairlineToReferenceBox = function(image, points) {
+  const box=window.hairlineReferenceBox(points);
+  const width=image.naturalWidth||image.width,height=image.naturalHeight||image.height;
+  if(!width||!height)throw new Error("发际线图片尚未加载");
+  const canvas=document.createElement("canvas");canvas.width=width;canvas.height=height;
+  const ctx=canvas.getContext("2d");
+  ctx.beginPath();box.corners.forEach((p,i)=>i?ctx.lineTo(...p):ctx.moveTo(...p));ctx.closePath();
+  ctx.clip();ctx.drawImage(image,0,0,width,height);
+  return canvas;
+};
+
+window.drawHairlineReferenceBox = function(ctx, points, crop, width, height) {
+  const box=window.hairlineReferenceBox(points);
+  const [x0,y0,x1,y1]=crop;
+  const screen=p=>[(p[0]-x0)/(x1-x0)*width,(p[1]-y0)/(y1-y0)*height];
+  const color="#e33399";
+  ctx.save();ctx.strokeStyle=color;ctx.lineWidth=Math.max(2.5,width/280);ctx.lineJoin="round";
+  ctx.beginPath();box.corners.forEach((p,i)=>{const q=screen(p);i?ctx.lineTo(...q):ctx.moveTo(...q);});ctx.closePath();ctx.stroke();
+  ctx.lineWidth=1.4;ctx.setLineDash([7,6]);ctx.beginPath();ctx.moveTo(...screen(box.landmarks[9]));ctx.lineTo(...screen(box.landmarks[152]));ctx.stroke();ctx.setLineDash([]);
+  const fontSize=Math.max(13,width/43);ctx.font=`600 ${fontSize}px sans-serif`;
+  for(const id of [9,152,162,389]){
+    const [x,y]=screen(box.landmarks[id]);
+    if(x<0||x>width||y<0||y>height)continue;
+    ctx.fillStyle=color;ctx.beginPath();ctx.arc(x,y,Math.max(3,width/160),0,Math.PI*2);ctx.fill();
+    const label=String(id),tw=ctx.measureText(label).width;
+    const tx=Math.max(3,Math.min(width-tw-8,x+7)),ty=Math.max(fontSize+4,Math.min(height-5,y-7));
+    ctx.fillStyle="rgba(255,255,255,.9)";ctx.fillRect(tx-3,ty-fontSize,tw+6,fontSize+4);ctx.fillStyle="#a51868";ctx.fillText(label,tx,ty);
+  }
+  const top=screen([(box.corners[2][0]+box.corners[3][0])/2,(box.corners[2][1]+box.corners[3][1])/2]);
+  const text="70% · 9→152",tw=ctx.measureText(text).width;
+  const tx=Math.max(4,Math.min(width-tw-6,top[0]-tw/2)),ty=Math.max(fontSize+4,top[1]-8);
+  ctx.fillStyle="rgba(255,255,255,.9)";ctx.fillRect(tx-3,ty-fontSize,tw+6,fontSize+4);ctx.fillStyle="#a51868";ctx.fillText(text,tx,ty);
+  ctx.restore();
+};
@@ -56,6 +56,7 @@
    <a href="#if3">接口3</a>
    <a href="#if4">接口4</a>
    <a href="#if5">接口5</a>
+    <a href="#if52">接口52</a>
    <a href="#if6">接口6</a>
    <a href="#errors">错误码</a>
    <a href="#test">在线测试</a>
@@ -106,6 +107,7 @@
        <tr><td><code>four_courts</code></td><td>object</td><td>四庭：top/upper/middle/lower，各含 <code>_cm</code> 和 <code>ratios</code></td></tr>
        <tr><td><code>seven_eyes</code></td><td>object</td><td>七眼：<code>eye_width_cm</code>/<code>face_width_cm</code>/<code>inter_eye_distance_cm</code> + <code>ratios</code> + <code>eye1</code>~<code>eye7</code>（从左到右 7 段宽度 cm，<code>eye1</code>/<code>eye7</code> 耳朵不可见时为 null）</td></tr>
        <tr><td><code>landmarks</code></td><td>object</td><td>5 个关键点像素坐标：hair_top/hairline/brow_center/nose_bottom/chin_tip</td></tr>
+        <tr><td><code>vertical_lines</code></td><td>object</td><td>标注图竖线坐标，与标注 PNG 同一套几何。原图像素：<code>y_top</code> / <code>y_bottom</code> 为竖线共用上下端（头顶之上到下巴之下，两端各伸出约短边的 3%，可能略超出图片）；<code>lines</code> 从左到右，每项 <code>{ name, x }</code>。name：<code>left_head_edge</code>（人头最左，耳朵不可见则不出现，此时 eye1 为 null）、<code>left_cheek</code>、<code>left_eye_outer</code>、<code>left_eye_inner</code>、<code>right_eye_inner</code>、<code>right_eye_outer</code>、<code>right_cheek</code>、<code>right_head_edge</code>（人头最右，不可见则不出现，此时 eye7 为 null）</td></tr>
        <tr><td><code>hairline_source</code></td><td>string</td><td>发际线来源：<code>"segmentation"</code>（真实分割，可信度高）/ <code>"estimated"</code>（比例估算，可信度低）</td></tr>
        <tr><td><code>head_pose</code></td><td>object</td><td>头部姿态角度：<code>{ yaw, pitch, roll }</code>（度），接近 0 表示正面照</td></tr>
        <tr><td><code>left_position</code></td><td>object</td><td>MediaPipe 21 号关键点坐标（左脸定位点），原图像素：<code>{ x: number, y: number }</code></td></tr>
@@ -126,6 +128,7 @@ const { code, data } = await res.json();
 // data.annotated_image_url  →  标注 PNG
 // data.face_total_height_cm  →  全脸高度
 // data.four_courts.ratios    →  { top_court: 0.25, ... }
+// data.vertical_lines       →  { y_top, y_bottom, lines: [{ name, x }, ...] }
 // 前端叠加显示：原图 + annotated_image_url（absolute 叠加）</pre>
    </div>
  </div>
@@ -251,6 +254,7 @@ console.log(features.gene_style);   // "自然型"</pre>
        <tr><td><code>four_courts</code></td><td>object</td><td>四庭：top/upper/middle/lower，各含 <code>_cm</code> 和 <code>ratios</code></td></tr>
        <tr><td><code>seven_eyes</code></td><td>object</td><td>七眼：<code>eye_width_cm</code>/<code>face_width_cm</code>/<code>inter_eye_distance_cm</code> + <code>ratios</code> + <code>eye1</code>~<code>eye7</code>（从左到右 7 段宽度 cm，<code>eye1</code>/<code>eye7</code> 耳朵不可见时为 null）</td></tr>
        <tr><td><code>landmarks</code></td><td>object</td><td>5 个关键点像素坐标：hair_top/hairline/brow_center/nose_bottom/chin_tip</td></tr>
+        <tr><td><code>vertical_lines</code></td><td>object</td><td>竖线坐标，结构同接口1</td></tr>
        <tr><td><code>hairline_source</code></td><td>string</td><td>发际线来源：<code>"segmentation"</code>（真实分割）/ <code>"estimated"</code>（比例估算）</td></tr>
        <tr><td><code>head_pose</code></td><td>object</td><td>头部姿态角度：<code>{ yaw, pitch, roll }</code>（度）</td></tr>
        <tr><td><code>left_position</code></td><td>object</td><td>MediaPipe 21 号关键点坐标（左脸定位点），原图像素：<code>{ x: number, y: number }</code></td></tr>
@@ -261,6 +265,16 @@ console.log(features.gene_style);   // "自然型"</pre>
    </div>
  </div>

+  <div class="card" id="if52">
+    <h2>52. 发际线新规则与裁切 &nbsp;<span class="badge post">POST</span> &nbsp;<code>/api/v1/hairline/generate-52</code></h2>
+    <div class="card-body">
+      <p class="desc">仅部署在GPU worker（8187），本次不新增网关路由。中下庭均值定位 middle，支持9→152扶正、人像裁切和矩形发际线预览。接口5保留原有三档实现；新功能使用此独立接口。</p>
+      <p>基础图片参数、<code>gender</code>、<code>hair_style</code>、<code>use_mask</code>、<code>prompt</code>、<code>generate_grow_image</code>沿用接口5。新增可选<code>normalize_image</code>，默认false，传true开启扶正与裁切；测试页显式开启。</p>
+      <p style="margin-top:12px">图片返回<code>*_base64</code>，空图片保留字段并返回null。另有<code>height_rule</code>、<code>framing</code>、<code>guide_landmarks</code>、原图及处理后底图。middle和坐标使用当前底图，high/low字段为null。标准裁切高约3.3333L、宽2.5L；长图在保留人脸的前提下裁为3:4。</p>
+      <p style="margin-top:12px"><a href="http://localhost:8187/static/test_interface52.html" class="link">52号规则测试页</a> · <a href="http://localhost:8187/static/test_interface52_compare.html" class="link">52号30案例对比页</a></p>
+    </div>
+  </div>
+
  <!-- ======== 接口6 ======== -->
  <div class="card" id="if6">
    <h2>6. 四庭七眼测量 v2 &nbsp;<span class="badge post">POST</span> &nbsp;<code>/api/v1/face/measure-v2</code> &nbsp;<span class="badge warn">v2</span></h2>
@@ -286,6 +300,7 @@ console.log(features.gene_style);   // "自然型"</pre>
        <tr><td><code>four_courts</code></td><td>object</td><td>三庭：upper/middle/lower，各含 _cm 和 ratios（<strong>无 top_court</strong>）</td></tr>
        <tr><td><code>seven_eyes</code></td><td>object</td><td>七眼：<code>eye_width_cm</code>/<code>face_width_cm</code>/<code>inter_eye_distance_cm</code> + <code>ratios</code> + <strong><code>eye2</code>~<code>eye6</code></strong>（左脸颊/左眼/两眼间距/右眼/右脸颊，5 段宽度 cm；<strong>无 eye1/eye7</strong>）</td></tr>
        <tr><td><code>landmarks</code></td><td>object</td><td>4 个关键点：hairline/brow_center/nose_bottom/chin_tip（<strong>无 hair_top</strong>）</td></tr>
+        <tr><td><code>vertical_lines</code></td><td>object</td><td>竖线坐标。结构同接口1，但 <code>y_top</code> / <code>y_bottom</code> 为发际线到下巴尖（不伸出），<code>lines</code> 只有中间 6 条，无 <code>left_head_edge</code> / <code>right_head_edge</code></td></tr>
        <tr><td><code>left_position</code></td><td>object</td><td>MediaPipe 21 号关键点坐标（左脸定位点），原图像素：<code>{ x: number, y: number }</code></td></tr>
        <tr><td><code>right_position</code></td><td>object</td><td>MediaPipe 251 号关键点坐标（右脸定位点，与 21 号镜像），原图像素：<code>{ x: number, y: number }</code></td></tr>
      </table>
@@ -323,6 +338,8 @@ console.log(features.gene_style);   // "自然型"</pre>
        <tr><td>3. B端生发</td><td><a href="/static/test_interface3.html" class="link">/static/test_interface3.html</a></td><td>划线图上传 → 生发效果图</td></tr>
        <tr><td>4. 用户特征</td><td><a href="/static/test_interface4.html" class="link">/static/test_interface4.html</a></td><td>上传照片 → 6项面部特征 + 原始JSON</td></tr>
        <tr><td>5. 发际线PNG</td><td><a href="/static/test_interface5.html" class="link">/static/test_interface5.html</a></td><td>上传+性别 → 发际线方案+中心点坐标</td></tr>
+        <tr><td>52. 发际线新规则</td><td><a href="http://localhost:8187/static/test_interface52.html" class="link">/static/test_interface52.html</a></td><td>中下庭定位、扶正裁切、矩形发际线叠加及生发</td></tr>
+        <tr><td>52. 规则对比</td><td><a href="http://localhost:8187/static/test_interface52_compare.html" class="link">/static/test_interface52_compare.html</a></td><td>30个案例，原图与旧/新规则样式对照</td></tr>
        <tr><td>6. 四庭七眼 v2</td><td><a href="/static/test_interface6.html" class="link">/static/test_interface6.html</a></td><td>同接口1，去顶庭 · 竖线发际线→下巴 · 无头部端线</td></tr>
      </table>
      <p style="font-size:12px;color:#94a3b8;margin-top:12px">
@@ -0,0 +1,418 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>接口52 — 发际线PNG生成 测试页</title>
+<style>
+  * { box-sizing: border-box; margin: 0; padding: 0; }
+  body { font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif; background: #f5f5f5; color: #333; }
+  .container { max-width: 1200px; margin: 0 auto; padding: 24px; }
+  h1 { font-size: 22px; margin-bottom: 6px; }
+  .subtitle { color: #888; font-size: 13px; margin-bottom: 24px; }
+
+  .card { background: #fff; border-radius: 12px; box-shadow: 0 1px 4px rgba(0,0,0,.06); margin-bottom: 20px; }
+  .card-header { font-weight: 700; font-size: 14px; padding: 14px 18px; border-bottom: 1px solid #f0f0f0; background: #fafafa; display: flex; justify-content: space-between; align-items: center; }
+  .card-body { padding: 18px; }
+
+  .upload-row { display: flex; gap: 12px; align-items: center; flex-wrap: wrap; }
+  .file-input { flex: 1; min-width: 200px; }
+  .file-input input[type=file] { width: 100%; padding: 8px; border: 2px dashed #ddd; border-radius: 8px; cursor: pointer; }
+  .form-group { display: flex; align-items: center; gap: 6px; white-space: nowrap; }
+  .form-group label { font-size: 13px; font-weight: 600; color: #374151; }
+  .form-group select { padding: 8px 12px; border: 1px solid #d1d5db; border-radius: 6px; font-size: 14px; }
+  .hair-checkboxes { display: flex; flex-wrap: wrap; gap: 4px 14px; align-items: center; }
+  .hair-checkboxes label { font-size: 13px; font-weight: 500; color: #374151; cursor: pointer; display: inline-flex; align-items: center; gap: 3px; white-space: nowrap; }
+  .hair-checkboxes label input[type=checkbox] { cursor: pointer; }
+  .checkbox-actions { display: flex; gap: 6px; margin-left: 4px; }
+  .checkbox-actions button { font-size: 11px; padding: 2px 8px; border-radius: 4px; border: 1px solid #d1d5db; background: #f9fafb; cursor: pointer; }
+  .checkbox-actions button:hover { background: #e5e7eb; }
+  .btn { padding: 10px 28px; border: none; border-radius: 8px; font-size: 15px; cursor: pointer; font-weight: 600; transition: .2s; }
+  .btn-primary { background: #2563eb; color: #fff; }
+  .btn-primary:hover { background: #1d4ed8; }
+  .btn-primary:disabled { background: #93c5fd; cursor: not-allowed; }
+  .btn-sm { padding: 6px 14px; font-size: 13px; }
+  .btn-outline { background: #fff; border: 1px solid #d1d5db; color: #374151; }
+  .btn-outline:hover { background: #f9fafb; }
+  .hint { font-size: 12px; color: #9ca3af; margin-top: 8px; }
+
+  .status { padding: 10px 16px; border-radius: 8px; font-size: 14px; margin-bottom: 16px; display: none; }
+  .status.info { background: #dbeafe; color: #1e40af; display: block; }
+  .status.error { background: #fee2e2; color: #991b1b; display: block; }
+  .status.success { background: #d1fae5; color: #065f46; display: block; }
+
+  .results-layout { display: flex; gap: 24px; }
+  .col-main { flex: 1.5; min-width: 0; }
+  .col-side { flex: 1; min-width: 0; }
+
+  /* 每个发型区块 */
+  .hair-block { padding: 16px 0; border-bottom: 1px solid #f0f0f0; }
+  .hair-block:first-child { padding-top: 4px; }
+  .hair-block:last-child { border-bottom: none; padding-bottom: 4px; }
+  .hair-block-title { font-size: 15px; font-weight: 700; color: #111827; margin-bottom: 10px; }
+
+  /* 未裁切原图 + middle + 矩形裁切发际线叠加 + 生发图 */
+  .level-row { display: grid; grid-template-columns: repeat(4, minmax(0, 1fr)); gap: 10px; align-items: start; }
+  .level-cell { background: #222; border-radius: 8px; overflow: hidden; display: flex; flex-direction: column; }
+  .level-cell .level-label { font-size: 12px; font-weight: 600; color: #fff; background: rgba(0,0,0,.55); padding: 4px 8px; min-height: 38px; display: flex; align-items: center; }
+  .level-cell img { width: 100%; max-height: 380px; object-fit: contain; display: block; }
+  .level-cell .level-none { color: #9ca3af; font-size: 13px; padding: 28px 0; text-align: center; }
+  /* 处理后底图 + 同尺寸透明设计曲线叠加 */
+  .level-cell .img-stack { position: relative; line-height: 0; }
+  .level-cell .img-stack .layer-base { width: 100%; max-height: 380px; object-fit: contain; display: block; }
+  .level-cell .img-stack .layer-anno { position: absolute; top: 0; left: 0; width: 100%; height: 100%; object-fit: contain; }
+  .hairline-crop-border { position: absolute; inset: 0; width: 100%; height: 100%; pointer-events: none; overflow: visible; }
+  .hairline-crop-border polygon { fill: none; stroke: #e33399; stroke-width: 2; stroke-dasharray: 6 4; stroke-linejoin: round; vector-effect: non-scaling-stroke; }
+  @media (max-width: 700px) { .level-row { grid-template-columns: repeat(2, 1fr); } }
+
+  /* 坐标高亮 */
+  .coord-box { background: #f9fafb; border-radius: 8px; padding: 12px 16px; margin-top: 10px; font-size: 14px; }
+  .coord-box .label { font-size: 11px; color: #9ca3af; margin-bottom: 4px; }
+  .coord-box .value { font-weight: 700; font-size: 18px; color: #111827; }
+
+  /* 四庭七眼测量（face_measure） */
+  .fm-section { margin-bottom: 18px; }
+  .fm-section:last-child { margin-bottom: 0; }
+  .fm-title { font-size: 13px; font-weight: 700; color: #374151; margin-bottom: 8px; }
+  .fm-grid { display: grid; grid-template-columns: repeat(auto-fill, minmax(120px, 1fr)); gap: 8px; }
+  .fm-cell { background: #f9fafb; border-radius: 6px; padding: 8px 10px; border: 1px solid #f0f0f0; }
+  .fm-cell .k { font-size: 11px; color: #9ca3af; margin-bottom: 2px; }
+  .fm-cell .v { font-size: 15px; font-weight: 700; color: #111827; }
+  .fm-cell .v .sub { font-size: 11px; font-weight: 500; color: #6b7280; margin-left: 4px; }
+  .fm-cell.muted .v { color: #9ca3af; font-weight: 600; }
+  .fm-tag { display: inline-block; font-size: 11px; padding: 2px 8px; border-radius: 10px; margin-left: 6px; font-weight: 600; }
+  .fm-tag.ok { background: #d1fae5; color: #065f46; }
+  .fm-tag.warn { background: #fef3c7; color: #92400e; }
+
+  .json-panel { max-height: 550px; overflow: auto; }
+  .json-content { padding: 14px 16px; font-family: "SF Mono", "Fira Code", monospace; font-size: 12px; line-height: 1.6; white-space: pre-wrap; word-break: break-all; }
+
+  .hidden { display: none !important; }
+  .rule-banner { background:#ecfdf5; border:1px solid #a7f3d0; border-radius:12px; padding:18px; margin-bottom:20px; line-height:1.8; }
+  .rule-banner strong { color:#065f46; font-size:19px; }
+  .rule-banner a { color:#065f46; margin-left:18px; font-weight:600; }
+  .placement { font-size:12px; color:#64748b; font-weight:400; margin:6px 0 10px; }
+  @media (max-width: 800px) { .results-layout { flex-direction: column; } }
+</style>
+<script src="/static/img_downscale.js?v=2"></script>
+<script src="/static/hairline_reference_box.js?v=2"></script>
+</head>
+<body>
+<div class="container">
+  <h1>💈 接口52 — 发际线 PNG 生成 测试</h1>
+  <p class="subtitle">上传正面照，选择性别与发型，查看按中下庭均值定位的设计曲线和生发效果。</p>
+  <div class="rule-banner"><strong>目标上庭高度 =（中庭高度 + 下庭高度）÷ 2</strong><a href="/static/test_interface52_compare.html">查看旧 / 新规则对比 →</a><br>上传图先扶正9→152，再裁切为高约 3.3333×L、宽 2.5×L，人物比原3L规则缩小约10%；L 是9号点到152号点的像素距离，眉心位置沿用参考图比例。长图会在保留人脸的前提下裁为3:4，其余空间不足时仅扶正。从眉心向上定位，各发型中点对齐同一目标。high、low 已停用，旧字段返回 null。</div>
+
+  <div class="card">
+    <div class="card-body">
+      <div class="upload-row">
+        <div class="file-input"><input type="file" id="imageFile" accept="image/jpeg,image/png,.jpg,.jpeg,.png"></div>
+        <div class="form-group">
+          <label>性别</label>
+          <select id="gender" onchange="onGenderChange()"><option value="female" selected>👩 Female（7种）</option><option value="male">👨 Male（7种）</option></select>
+        </div>
+        <div class="form-group" style="align-items:flex-start">
+          <label style="padding-top:3px">发型 *</label>
+          <div>
+            <div class="hair-checkboxes" id="hairStyleGroup"></div>
+            <div class="checkbox-actions" style="margin-top:4px">
+              <button type="button" onclick="selectAllHair(true)">全选</button>
+              <button type="button" onclick="selectAllHair(false)">全不选</button>
+            </div>
+          </div>
+        </div>
+        <div class="form-group">
+          <label>生发效果图</label>
+          <label class="checkbox-inline" style="font-weight:normal;display:flex;align-items:center;gap:6px">
+            <input type="checkbox" id="genGrowImg" checked> 同时生成生发效果（关闭后只查看设计曲线，速度更快）
+          </label>
+        </div>
+        <button class="btn btn-primary" id="submitBtn" onclick="submitTest()">🚀 提交</button>
+        <button class="btn btn-outline btn-sm" onclick="clearResults()">清除</button>
+      </div>
+      <div class="hint">JPG/PNG 正面照 &nbsp;|&nbsp; Female: 椭圆/花瓣/心形/直线/波浪/大花瓣/古典花瓣 &nbsp;|&nbsp; Male: 椭圆/倒弧/M形/直线/桃心/柔和花瓣/波浪 &nbsp;|&nbsp; 每个发型返回一张设计曲线，生发遮罩同步到新位置</div>
+      <div id="statusBar" class="status hidden"></div>
+    </div>
+  </div>
+
+  <div class="results-layout hidden" id="resultsArea">
+    <div class="col-main">
+      <!-- 四庭七眼测量（face_measure，复用接口1数值） -->
+      <div class="card">
+        <div class="card-header"><span>📐 处理后图像的四庭七眼测量</span><span style="font-weight:400;font-size:12px;color:#9ca3af">与当前显示底图使用同一坐标系</span></div>
+        <div class="card-body"><div id="faceMeasureArea"><span style="color:#9ca3af">—</span></div></div>
+      </div>
+      <!-- 所有发型平铺 -->
+      <div class="card">
+        <div class="card-header"><span>🎯 未裁切原图 + middle + 裁切发际线叠加 + 生发图</span><span style="font-weight:400;font-size:12px;color:#9ca3af" id="resultCount">共 0 个发型</span></div>
+        <div class="card-body">
+          <div class="coord-box" style="margin-top:0;margin-bottom:16px">
+            <div class="label">📍 设计中心点 — 图片像素坐标</div>
+            <div class="value">
+              <span id="centerPoint">—</span>
+            </div>
+          </div>
+          <div id="ruleDetails" style="font-size:13px;color:#475569;margin-bottom:16px;line-height:1.8"></div>
+          <div id="resultsGrid"></div>
+        </div>
+      </div>
+    </div>
+    <div class="col-side">
+      <div class="card">
+        <div class="card-header"><span>📋 JSON</span><button class="btn btn-outline btn-sm" onclick="copyJson()">📋 复制</button></div>
+        <div class="json-panel"><pre class="json-content" id="jsonContent"></pre></div>
+      </div>
+    </div>
+  </div>
+</div>
+
+<script>
+const API_BASE = window.location.origin;
+let _images = [], _uploadUrl = '', _originalSrc = '', _processedSrc = '', _center = null, _guideLandmarks = null, _renderVersion = 0;
+
+function $(id) { return document.getElementById(id); }
+function setStatus(t, type) { const b=$('statusBar'); b.textContent=t; b.className='status '+type; }
+
+// 发型列表（同接口2）
+const HAIR_STYLES = {
+  female: [
+    { value:'1', label:'1. ellipse 椭圆' },
+    { value:'2', label:'2. flower 花瓣' },
+    { value:'3', label:'3. heart 心形' },
+    { value:'4', label:'4. straight 直线' },
+    { value:'5', label:'5. wave 波浪' },
+    { value:'6', label:'6. bigflower 大花瓣型' },
+    { value:'7', label:'7. clasicalflower 古典花瓣型' },
+  ],
+  male: [
+    { value:'1', label:'1. ellipse 椭圆' },
+    { value:'2', label:'2. inverse_arc 倒弧' },
+    { value:'3', label:'3. m M形' },
+    { value:'4', label:'4. straight 直线' },
+    { value:'5', label:'5. heart 桃心形' },
+    { value:'6', label:'6. Softpetal 柔和花瓣形' },
+    { value:'7', label:'7. wave 波浪形' },
+  ]
+};
+
+function onGenderChange() {
+  const g = $('gender').value;
+  const div = $('hairStyleGroup');
+  div.innerHTML = '';
+  HAIR_STYLES[g].forEach(s => {
+    const cb = document.createElement('input');
+    cb.type = 'checkbox'; cb.value = s.value; cb.id = 'hs_' + s.value;
+    const lb = document.createElement('label');
+    lb.htmlFor = cb.id; lb.appendChild(cb);
+    lb.appendChild(document.createTextNode(' ' + s.label));
+    div.appendChild(lb);
+  });
+}
+
+function selectAllHair(select) {
+  document.querySelectorAll('#hairStyleGroup input[type=checkbox]').forEach(cb => { cb.checked = select; });
+}
+
+async function submitTest() {
+  let f = $('imageFile').files[0];
+  if (!f) { setStatus('请选择图片', 'error'); return; }
+  const checked = [...document.querySelectorAll('#hairStyleGroup input:checked')].map(cb => cb.value);
+  if (!checked.length) { setStatus('请至少选择一个发型', 'error'); return; }
+  // Preserve native pixels: the worker frames and straightens the image first.
+
+  if (_uploadUrl) URL.revokeObjectURL(_uploadUrl);
+  _uploadUrl = URL.createObjectURL(f);
+  _originalSrc = _uploadUrl; _processedSrc = ''; _images = []; _guideLandmarks = null; _renderVersion++;
+  $('resultsGrid').innerHTML = '';
+  $('submitBtn').disabled = true; $('submitBtn').textContent = '⏳ ...';
+  setStatus('请求中...', 'info'); $('resultsArea').classList.add('hidden');
+
+  const fd = new FormData(); fd.append('image_file', f); fd.append('gender', $('gender').value); fd.append('hair_style', checked.join(',')); fd.append('generate_grow_image', $('genGrowImg').checked ? 'true' : 'false'); fd.append('normalize_image', 'true');
+  const _reqStart = performance.now();
+  try {
+    const r = await fetch(API_BASE + '/api/v1/hairline/generate-52', { method:'POST', headers: { 'X-Internal-Token': 'dev-shared-secret-2026' }, body:fd });
+    const json = await r.json();
+    const _elapsed = ((performance.now() - _reqStart) / 1000).toFixed(2);
+    $('jsonContent').textContent = JSON.stringify(json, null, 2);
+    $('resultsArea').classList.remove('hidden');
+    if (json.code === 0) {
+      _images = json.data.hairline_images || [];
+      _originalSrc = resolveImgSrc(json.data.original_image_url, json.data.original_image_base64, 'image/jpeg') || _uploadUrl;
+      _processedSrc = resolveImgSrc(json.data.input_image_url, json.data.input_image_base64, 'image/jpeg');
+      _guideLandmarks = json.data.guide_landmarks;
+      if (_originalSrc !== _uploadUrl) { URL.revokeObjectURL(_uploadUrl); _uploadUrl = ''; }
+      _center = json.data.best_hairline_center_point;
+      const _fmtPt = p => p ? '(' + p.x + ', ' + p.y + ')' : '—';
+      $('centerPoint').textContent = _fmtPt(json.data.best_hairline_center_point);
+      const rule = json.data.height_rule;
+      $('ruleDetails').textContent = rule
+        ? '中庭 ' + rule.middle_height_px.toFixed(2) + ' px + 下庭 ' + rule.lower_height_px.toFixed(2) + ' px → 目标上庭 ' + rule.target_upper_height_px.toFixed(2) + ' px' + (rule.target_upper_height_cm == null ? '' : '（约 ' + rule.target_upper_height_cm.toFixed(2) + ' cm）') + '；目标 y = ' + rule.target_center.y.toFixed(2) + '。测量来源：' + (rule.measurement_source === 'face_measure' ? '当前底图测量' : '发际线关键点')
+        : '—';
+      if (json.data.framing) { const f = json.data.framing; $('ruleDetails').textContent += '；' + f.message + '，旋转 ' + f.rotation_degrees.toFixed(3) + '°，底图 ' + f.output_size.width + '×' + f.output_size.height + '。'; }
+      $('resultCount').textContent = '共 ' + _images.length + ' 个发型';
+      setStatus('✅ ' + _images.length + ' 个发型已按新规则定位 (' + _elapsed + 's)', 'success');
+      await renderGrid();
+      renderFaceMeasure(json.data.face_measure);
+    } else {
+      setStatus('❌ (' + _elapsed + 's) code=' + json.code + ' ' + json.message, 'error');
+    }
+  } catch(e) {
+    const _elapsed = ((performance.now() - _reqStart) / 1000).toFixed(2);
+    setStatus('❌ (' + _elapsed + 's) ' + e.message, 'error'); $('resultsArea').classList.remove('hidden'); }
+  finally { $('submitBtn').disabled = false; $('submitBtn').textContent = '🚀 提交'; }
+}
+
+async function buildClippedOverlay(src, points) {
+  const image = await new Promise((resolve, reject) => {
+    const img = new Image(); img.crossOrigin = 'anonymous';
+    img.onload = () => resolve(img); img.onerror = () => reject(new Error('无法读取发际线图片'));
+    img.src = src;
+  });
+  const canvas = window.clipHairlineToReferenceBox(image, points);
+  const corners = window.hairlineReferenceBox(points).corners.map(p => p.join(',')).join(' ');
+  const result = {
+    src: canvas.toDataURL('image/png'),
+    border: '<svg class="hairline-crop-border" viewBox="0 0 '+canvas.width+' '+canvas.height+'" role="img" aria-label="粉色虚线：发际线裁切边界"><polygon points="'+corners+'"></polygon></svg>'
+  };
+  canvas.width = 0; canvas.height = 0;
+  return result;
+}
+
+async function renderGrid() {
+  const version = ++_renderVersion;
+  if (!_images.length) { $('resultsGrid').innerHTML = '<span style="color:#9ca3af">无数据</span>'; return; }
+  const images = _images, originalSrc = _originalSrc, processedSrc = _processedSrc, guideLandmarks = _guideLandmarks;
+  const escapeText = text => String(text).replace(/[&<>"']/g, c => ({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c]));
+  // The overlay uses the server's processed image, with matching coordinates.
+  const overlayCell = (label, src, className, emptyText = '暂无处理后图片', border = '') => (src && processedSrc)
+    ? '<div class="level-cell '+className+'"><div class="level-label">'+label+'</div><div class="img-stack">'+
+        '<img class="layer-base" src="'+processedSrc+'" alt="处理后底图">'+
+        '<img class="layer-anno" src="'+src+'" alt="'+label+'">'+border+'</div></div>'
+    : '<div class="level-cell '+className+'"><div class="level-label">'+label+'</div><div class="level-none">'+escapeText(emptyText)+'</div></div>';
+  const photoCell = (label, src, className) => src
+    ? '<div class="level-cell '+className+'"><div class="level-label">'+label+'</div><img src="'+src+'" alt="'+label+'"></div>'
+    : '<div class="level-cell '+className+'"><div class="level-label">'+label+'</div><div class="level-none">无</div></div>';
+  const blocks = await Promise.all(images.map(async function(it) {
+    // 兼容网关 *_url 与 worker *_base64
+    const mid = resolveImgSrc(it.image_middle_url, it.image_middle_base64, 'image/png');
+    const grown = resolveImgSrc(it.grown_image_url, it.grown_image_base64, 'image/jpeg');
+    let clipped = '', clipBorder = '', clipError = '缺少发际线图片或参考点';
+    if (mid && processedSrc && guideLandmarks) {
+      try { const preview = await buildClippedOverlay(mid, guideLandmarks); clipped = preview.src; clipBorder = preview.border; }
+      catch (error) { clipError = error.message; }
+    }
+    return '<div class="hair-block">' +
+      '<div class="hair-block-title">#' + (it.order||'—') + ' ' + (it.hairline_type||'') +
+      (grown ? ' &nbsp;<span style="font-size:11px;color:#7c3aed">含生发图</span>' : '') + '</div>' +
+      (it.height_placement ? '<div class="placement">中心 y：' + it.height_placement.new_center.y.toFixed(2) + ' · 相对旧规则 ' + (it.height_placement.shift_y_px >= 0 ? '下移 ' : '上移 ') + Math.abs(it.height_placement.shift_y_px).toFixed(2) + ' px · 目标误差 ' + it.height_placement.target_error_px.toFixed(2) + ' px</div>' : '') +
+      '<div class="level-row">' +
+        photoCell('未裁切原图', originalSrc, 'original-cell') +
+        overlayCell('middle · 发际线叠加', mid, 'middle-cell') +
+        overlayCell('裁切发际线叠加', clipped, 'clipped-cell', clipError, clipBorder) +
+        photoCell('生发图', grown, 'grown-cell') +
+      '</div>' +
+    '</div>';
+  }));
+  if (version === _renderVersion) $('resultsGrid').innerHTML = blocks.join('');
+}
+
+function clearResults() {
+  if (_uploadUrl) URL.revokeObjectURL(_uploadUrl);
+  _images=[]; _uploadUrl=''; _originalSrc=''; _processedSrc=''; _center=null; _guideLandmarks=null; _renderVersion++;
+  $('ruleDetails').textContent = '';
+  $('resultsArea').classList.add('hidden'); $('statusBar').className='status hidden';
+  $('imageFile').value=''; $('jsonContent').textContent=''; $('resultsGrid').innerHTML='';
+  $('centerPoint').textContent='—';
+  $('resultCount').textContent = '共 0 个发型';
+  $('faceMeasureArea').innerHTML = '<span style="color:#9ca3af">—</span>';
+}
+
+// 四庭七眼测量渲染（face_measure，接口1数值）
+const _PCT = r => (r == null ? '' : '（' + (r * 100).toFixed(1) + '%）');
+const _FMT = v => (v == null ? '—' : Number(v).toFixed(2));
+function renderFaceMeasure(fm) {
+  const box = $('faceMeasureArea');
+  if (!fm) {
+    box.innerHTML = '<div style="padding:24px;text-align:center;color:#9ca3af;font-size:14px">⚠️ face_measure = null（测量失败或未启用）</div>';
+    return;
+  }
+  // 发际线来源 + 头部姿态
+  const src = fm.hairline_source;
+  const srcTag = src === 'segmentation'
+    ? '<span class="fm-tag ok">segmentation 真实分割</span>'
+    : (src === 'estimated' ? '<span class="fm-tag warn">estimated 比例估算</span>' : '');
+  let poseHtml = '—';
+  if (fm.head_pose) {
+    const p = fm.head_pose;
+    poseHtml = 'yaw ' + _FMT(p.yaw) + '° &nbsp; pitch ' + _FMT(p.pitch) + '° &nbsp; roll ' + _FMT(p.roll) + '°';
+  }
+  let h = '<div class="coord-box" style="margin-top:0;margin-bottom:16px;display:flex;gap:32px;flex-wrap:wrap;align-items:center">'
+    + '<div><div class="label">脸总高</div><div class="value">' + _FMT(fm.face_total_height_cm) + ' cm</div></div>'
+    + '<div><div class="label">发际线来源</div><div class="value" style="font-size:15px">' + (src || '—') + srcTag + '</div></div>'
+    + '<div><div class="label">头部姿态</div><div class="value" style="font-size:14px">' + poseHtml + '</div></div>'
+    + '</div>';
+
+  // 四庭
+  const fc = fm.four_courts || {};
+  const fr = (fc.ratios || {});
+  const courts = [
+    ['顶庭', fc.top_court_cm, fr.top_court],
+    ['上庭', fc.upper_court_cm, fr.upper_court],
+    ['中庭', fc.middle_court_cm, fr.middle_court],
+    ['下庭', fc.lower_court_cm, fr.lower_court],
+  ];
+  h += '<div class="fm-section"><div class="fm-title">📏 四庭（纵向，自上而下）</div><div class="fm-grid">';
+  courts.forEach(c => {
+    h += '<div class="fm-cell"><div class="k">' + c[0] + '</div><div class="v">' + _FMT(c[1]) + ' cm<span class="sub">' + _PCT(c[2]).replace(/[（）]/g, '') + '</span></div></div>';
+  });
+  h += '</div></div>';
+
+  // 七眼
+  const se = fm.seven_eyes || {};
+  const ser = (se.ratios || {});
+  h += '<div class="fm-section"><div class="fm-title">👁 七眼（横向汇总）</div><div class="fm-grid">';
+  [['单眼宽', se.eye_width_cm, ser.eye_width], ['脸宽', se.face_width_cm, null], ['两眼间距', se.inter_eye_distance_cm, ser.inter_eye_distance]].forEach(e => {
+    h += '<div class="fm-cell"><div class="k">' + e[0] + '</div><div class="v">' + _FMT(e[1]) + ' cm' + (e[2] != null ? '<span class="sub">' + (e[2] * 100).toFixed(1) + '%</span>' : '') + '</div></div>';
+  });
+  h += '</div></div>';
+
+  // 七眼 7 段（从左到右）
+  const segLabels = ['eye1 左耳外段', 'eye2 左脸颊段', 'eye3 左眼', 'eye4 两眼间距', 'eye5 右眼', 'eye6 右脸颊段', 'eye7 右耳外段'];
+  h += '<div class="fm-section"><div class="fm-title">↔️ 七眼 7 段宽度（从左到右，cm）</div><div class="fm-grid" style="grid-template-columns:repeat(7,1fr)">';
+  segLabels.forEach((lab, i) => {
+    const v = se['eye' + (i + 1)];
+    const muted = v == null ? ' muted' : '';
+    h += '<div class="fm-cell' + muted + '"><div class="k">' + lab + '</div><div class="v">' + (v == null ? '—' : _FMT(v)) + '</div></div>';
+  });
+  h += '</div></div>';
+
+  // 关键点坐标
+  const lm = fm.landmarks || {};
+  const lmLabels = { hair_top: '头顶', hairline: '发际线', brow_center: '眉心', nose_bottom: '鼻翼下缘', chin_tip: '下巴尖' };
+  h += '<div class="fm-section"><div class="fm-title">📍 关键分界点（当前底图像素坐标）</div><div class="fm-grid">';
+  Object.keys(lmLabels).forEach(k => {
+    const p = lm[k];
+    h += '<div class="fm-cell"><div class="k">' + lmLabels[k] + '</div><div class="v" style="font-size:14px">' + (p ? '(' + p.x + ', ' + p.y + ')' : '—') + '</div></div>';
+  });
+  h += '</div></div>';
+
+  box.innerHTML = h;
+}
+
+function copyJson() {
+  navigator.clipboard.writeText($('jsonContent').textContent).then(() => {
+    const b = event.target, o = b.textContent; b.textContent = '✅ 已复制'; setTimeout(() => b.textContent = o, 1500);
+  });
+}
+
+document.addEventListener('DOMContentLoaded', () => {
+  onGenderChange();  // 初始填充发型列表
+  const dz = $('imageFile');
+  dz.addEventListener('dragover', e => { e.preventDefault(); dz.style.borderColor='#2563eb'; });
+  dz.addEventListener('dragleave', () => dz.style.borderColor='#ddd');
+  dz.addEventListener('drop', e => { e.preventDefault(); dz.style.borderColor='#ddd'; if(e.dataTransfer.files.length){ dz.files=e.dataTransfer.files; setStatus('已选择: '+e.dataTransfer.files[0].name, 'info'); } });
+  dz.addEventListener('change', () => { if(dz.files.length) setStatus('已选择: '+dz.files[0].name, 'info'); });
+});
+</script>
+</body>
+</html>
@@ -0,0 +1,52 @@
+<!doctype html>
+<html lang="zh-CN">
+<head>
+  <meta charset="utf-8">
+  <meta name="viewport" content="width=device-width,initial-scale=1">
+  <title>接口 52 · 发际线高度规则对比</title>
+  <link rel="stylesheet" href="/static/hairline_height_compare.css">
+  <script src="/static/img_downscale.js?v=2" defer></script>
+  <script src="/static/hairline_reference_box.js" defer></script>
+  <script src="/static/hairline_height_compare.js" defer></script>
+</head>
+<body>
+<main>
+  <header class="page-header">
+    <div><div class="eyebrow">接口 52 / 规则验证</div><h1>发际线高度规则对比</h1><p>用同一张照片、同一个形状，查看旧规则与中下庭均值规则的位置差异。</p></div>
+    <a class="button secondary" href="/static/test_interface52.html">打开规则测试页 ↗</a>
+  </header>
+  <section class="rule-card" aria-label="规则说明">
+    <div><span class="pill">新规则</span><h2>目标上庭 =（中庭 + 下庭）÷ 2</h2><p>从眉心向上定位，所有形状的中点对齐同一高度。各发型保留自身轮廓，叠加在扶正后的底图上。</p></div>
+    <div class="rule-note"><span class="dot old"></span>旧规则：根据头发边界和模板定位<br><span class="dot new"></span>新规则：根据中庭、下庭定位<br><span class="dash"></span>虚线：新规则的目标高度</div>
+  </section>
+  <details class="framing-standard"><summary>查看已扶正的标准图与裁切参数</summary><div id="framingStandard">正在加载标准图参数…</div></details>
+  <section class="stats" aria-label="验证概况">
+    <div><span>已验证图片</span><strong id="statCases">—</strong><small>独立测试图片</small></div>
+    <div><span>发际线对比</span><strong id="statComparisons">—</strong><small>男性 / 女性各 7 型</small></div>
+    <div><span>最大中心误差</span><strong id="statError">—</strong><small>新中心与目标的像素差</small></div>
+    <div><span>验证状态</span><strong id="statStatus">加载中</strong><small id="runTime">正在读取已保存结果</small></div>
+  </section>
+  <section class="toolbar" aria-label="显示设置">
+    <label>发型集合<select id="genderFilter"><option value="all" selected>全部集合（男女各 7 型）</option><option value="female">女性发际线</option><option value="male">男性发际线</option></select></label>
+    <label>形状<select id="styleFilter"></select></label>
+    <label>图片<select id="caseFilter"><option value="all">全部 30 个案例</option></select></label>
+    <div class="segmented" aria-label="对比方式"><button id="sideMode" class="active" type="button">左右对比</button><button id="overlayMode" type="button">同图叠线</button></div>
+    <label class="check"><input type="checkbox" id="zoomFace">额头特写</label>
+    <label class="check"><input type="checkbox" id="showTarget" checked>目标参考线</label>
+    <label class="check"><input type="checkbox" id="showBox" checked>新图参考框</label>
+  </section>
+  <p id="status" class="status" role="status" aria-live="polite">正在加载30个案例的对比结果…</p>
+  <section id="results" class="results" aria-label="案例对比"></section>
+  <section class="tools-card">
+    <div class="tools-heading"><div><h2>继续验证</h2><p>上传新图片，或重新运行清单全部30张。每人先显示未裁切原图，再显示旧、新规则；对比只生成定位曲线，无需等待生发。</p></div><button id="rerunBatch" class="button secondary" type="button">重跑全部30个案例</button></div>
+    <form id="customForm" class="custom-form">
+      <label>选择图片<input id="customFile" type="file" accept="image/png,image/jpeg" required></label>
+      <label>发型集合<select id="customGender"><option value="female">女性 · 7 型</option><option value="male">男性 · 7 型</option></select></label>
+      <button class="button primary" id="customRun" type="submit">运行自定义对比</button>
+    </form>
+    <p class="footnote">high、low 已停用；接口旧字段保留为 null。当前页的“旧规则”是本次高度修改前的定位逻辑。旧、新规则使用同一张处理后底图。</p>
+  </section>
+  <footer>旧线 <span class="dot old"></span> 橙色 · 新线 <span class="dot new"></span> 青色 · 粉色参考框：两侧均平行 9→152，左边经过 162、右边经过 389、底边经过 9；高度为 9→152 距离的 70%<br>向上移动为负，向下移动为正 · 所有距离基于接口实际输入图片</footer>
+</main>
+</body>
+</html>
@@ -0,0 +1,3 @@
+<!doctype html>
+<html lang="zh-CN"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>接口52 · 规则对比</title><meta http-equiv="refresh" content="0;url=/static/test_interface52_compare.html"></head>
+<body><p>新规则对比已迁移至接口52：<a href="/static/test_interface52_compare.html">打开接口52对比页</a></p><script>location.replace('/static/test_interface52_compare.html'+location.search+location.hash);</script></body></html>
@@ -0,0 +1,38 @@
+{
+  "description": "用户指定的接口 52 发际线高度规则回归验证图片清单",
+  "recorded_on": "2026-11-03",
+  "endpoint": "/api/v1/hairline/generate-52",
+  "total": 30,
+  "cases": [
+    {"image_number": 1, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%201.png"},
+    {"image_number": 2, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%202.png"},
+    {"image_number": 3, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%203.png"},
+    {"image_number": 4, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%204.png"},
+    {"image_number": 5, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%205.png"},
+    {"image_number": 6, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%206.png"},
+    {"image_number": 7, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%207.png"},
+    {"image_number": 20, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2020.png"},
+    {"image_number": 21, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2021.png"},
+    {"image_number": 22, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2022.png"},
+    {"image_number": 23, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2023.png"},
+    {"image_number": 27, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2027.png"},
+    {"image_number": 32, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2032.png"},
+    {"image_number": 33, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2033.png"},
+    {"image_number": 34, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2034.png"},
+    {"image_number": 35, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2035.png"},
+    {"image_number": 36, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2036.png"},
+    {"image_number": 37, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2037.png"},
+    {"image_number": 40, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2040.png"},
+    {"image_number": 41, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2041.png"},
+    {"image_number": 42, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2042.png"},
+    {"image_number": 43, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2043.png"},
+    {"image_number": 44, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2044.png"},
+    {"image_number": 45, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2045.png"},
+    {"image_number": 46, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2046.png"},
+    {"image_number": 47, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2047.png"},
+    {"image_number": 50, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2050.png"},
+    {"image_number": 51, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2051.png"},
+    {"image_number": 52, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2052.png"},
+    {"image_number": 53, "image_url": "https://raw.githubusercontent.com/feastfuyan/yonghe-badcase-0917/main/%E5%9B%BE%E7%89%87%2053.png"}
+  ]
+}
@@ -18,7 +18,7 @@ H = {"X-Internal-Token": "testpass"}
@pytest.fixture(scope="module")
 def client():
    # with 触发 lifespan：加载模型单例（detector + 尽力加载 segmenter）
-    with TestClient(app_module.app) as c:
+    with TestClient(app_module.app, client=("127.0.0.1", 50000)) as c:
        yield c


@@ -36,12 +36,76 @@ def test_auth_missing_token_401(client):
    assert r.status_code == 401


-def test_health_no_token_200(client):
+def _patch_health_ok(monkeypatch):
+    """让 /health 深检确定性通过：不依赖本机磁盘水位与 8188/8801 服务存活。"""
+    monkeypatch.setattr(app_module, "_HEALTH_CFG", {
+        **app_module._HEALTH_CFG, "min_free_gb": 0.0, "check_deps": True})
+
+    async def _deps_ok():
+        return True, {"comfyui": True, "change_hair": True}
+    monkeypatch.setattr(app_module, "_check_deps", _deps_ok)
+
+
+def test_health_no_token_200(client, monkeypatch):
+    _patch_health_ok(monkeypatch)
    r = client.get("/health")
    assert r.status_code == 200
    assert r.json()["status"] == "ok"


+def test_health_disk_low_503(client, monkeypatch):
+    # 阈值调到不可能的 99999G → 必然触发磁盘水位 503
+    monkeypatch.setattr(app_module, "_HEALTH_CFG", {
+        **app_module._HEALTH_CFG, "min_free_gb": 99999.0, "check_deps": False})
+    r = client.get("/health")
+    assert r.status_code == 503
+    body = r.json()
+    assert body["status"] == "degraded"
+    assert body["reason"].startswith("disk_free_below")
+
+
+def test_health_dep_down_503(client, monkeypatch):
+    monkeypatch.setattr(app_module, "_HEALTH_CFG", {
+        **app_module._HEALTH_CFG, "min_free_gb": 0.0, "check_deps": True})
+
+    async def _deps_down():
+        return False, {"comfyui": False, "change_hair": True}
+    monkeypatch.setattr(app_module, "_check_deps", _deps_down)
+    r = client.get("/health")
+    assert r.status_code == 503
+    body = r.json()
+    assert body["status"] == "degraded"
+    assert "comfyui" in body["reason"]
+
+
+def test_health_ok_carries_detail(client, monkeypatch):
+    _patch_health_ok(monkeypatch)
+    r = client.get("/health")
+    assert r.status_code == 200
+    body = r.json()
+    assert "disk_free_gb" in body
+    assert body["deps"] == {"comfyui": True, "change_hair": True}
+
+
+def test_tcp_probe_real_socket():
+    # 用进程内真实 socket 验证探测本身（曾因 open_connection 返回元组误判全部不可达）
+    import asyncio
+
+    async def scenario():
+        server = await asyncio.start_server(lambda r, w: None, host="127.0.0.1", port=0)
+        port = server.sockets[0].getsockname()[1]
+        up = await app_module._tcp_probe("127.0.0.1", port)
+        server.close()
+        await server.wait_closed()
+        # 端口 1 无监听，本机 connect 立即被拒
+        down = await app_module._tcp_probe("127.0.0.1", 1)
+        return up, down
+
+    up, down = asyncio.run(scenario())
+    assert up is True
+    assert down is False
+
+
 def test_param_none_provided_1007(client):
    r = client.post(URL, headers=H)
    assert r.json()["code"] == 1007
@@ -125,6 +189,40 @@ def test_growb_success(client, monkeypatch):
 HLGEN = "/api/v1/hairline/generate"


+def test_hairline5_restored_schema_has_only_historical_parameters(client):
+    schema = client.get("/openapi.json", headers=H).json()
+    ref = schema["paths"][HLGEN]["post"]["requestBody"]["content"]["multipart/form-data"]["schema"]["$ref"]
+    fields = schema["components"]["schemas"][ref.rsplit("/", 1)[-1]]["properties"]
+    assert set(fields) == {"image_file", "image_url", "image_base64", "gender", "hair_style",
+                           "use_mask", "prompt", "generate_grow_image"}
+    assert fields["use_mask"]["default"] is True
+    assert fields["generate_grow_image"]["default"] is True
+    assert fields["prompt"]["default"] == "填充遮罩区域的头发，人物增加一点磨皮"
+
+
+@pytest.mark.parametrize("gender", ["female", "male"])
+def test_hairline5_restored_three_levels_and_original_coordinate_contract(client, gender):
+    from pathlib import Path
+    photo = Path(fixture("frontal.jpg")).read_bytes()
+    body = client.post(HLGEN, headers=H, files={"image_file": ("frontal.jpg", photo, "image/jpeg")},
+                       data={"gender": gender, "hair_style": "1,3", "generate_grow_image": "false"}).json()
+    assert body["code"] == 0, body
+    data = body["data"]
+    assert set(data) == {"hairline_images", "best_hairline_center_point", "high_hairline_center_point",
+                         "low_hairline_center_point", "face_measure"}
+    expected = _cv2.imdecode(_np.frombuffer(photo, _np.uint8), _cv2.IMREAD_COLOR).shape[:2]
+    for key in ("best_hairline_center_point", "high_hairline_center_point", "low_hairline_center_point"):
+        assert set(data[key]) == {"x", "y"}
+        assert isinstance(data[key]["x"], int) and isinstance(data[key]["y"], int)
+    for item in data["hairline_images"]:
+        assert set(item) == {"hairline_type", "image_middle_base64", "image_high_base64", "image_low_base64",
+                             "grown_image_base64", "order"}
+        assert item["grown_image_base64"] is None
+        for level in ("middle", "high", "low"):
+            image = _cv2.imdecode(_np.frombuffer(base64.b64decode(item[f"image_{level}_base64"]), _np.uint8), _cv2.IMREAD_UNCHANGED)
+            assert image.shape == (*expected, 4)
+
+
 def test_hairline_gen_missing_gender_1004(client):
    files = {"image_file": ("frontal.jpg", open(fixture("frontal.jpg"), "rb"), "application/octet-stream")}
    assert client.post(HLGEN, headers=H, files=files).json()["code"] == 1004
@@ -177,6 +275,181 @@ def test_hairline_gen_female(client, monkeypatch):
    assert set(["yaw", "pitch", "roll"]).issubset(fm["head_pose"].keys())


+HLGEN52 = "/api/v1/hairline/generate-52"
+
+
+def test_hairline52_legacy_form_parameters_and_defaults_are_preserved(client):
+    schema = client.get("/openapi.json", headers=H).json()
+    body = schema["paths"][HLGEN52]["post"]["requestBody"]["content"]["multipart/form-data"]["schema"]
+    body = schema["components"]["schemas"][body["$ref"].rsplit("/", 1)[-1]]
+    fields = body["properties"]
+    legacy = {"image_file", "image_url", "image_base64", "gender", "hair_style",
+              "use_mask", "prompt", "generate_grow_image"}
+    assert legacy.issubset(fields)
+    assert not set(body.get("required", [])) - legacy
+    for key in ("image_file", "image_url", "image_base64", "gender", "hair_style"):
+        assert any(choice.get("type") == "string" for choice in fields[key].get("anyOf", [fields[key]]))
+    assert fields["use_mask"]["type"] == "boolean" and fields["use_mask"]["default"] is True
+    assert fields["generate_grow_image"]["type"] == "boolean" and fields["generate_grow_image"]["default"] is True
+    assert fields["prompt"]["default"] == "填充遮罩区域的头发，人物增加一点磨皮"
+    assert fields["normalize_image"]["default"] is False
+
+
+@pytest.mark.parametrize("image_kind", ["file", "base64", "url"])
+def test_hairline52_legacy_request_keeps_fields_and_original_coordinates(client, monkeypatch, image_kind):
+    import contextlib
+    from pathlib import Path
+    from types import SimpleNamespace
+    import httpx
+    import hairline.comfyui as comfy
+
+    raw = Path(fixture("frontal.jpg")).read_bytes()
+    calls = []
+    def generate(*args, **kwargs):
+        calls.append(kwargs)
+        return _PNG_1x1
+    monkeypatch.setattr(comfy, "run", generate)
+    form = {"gender": "male", "hair_style": "1", "use_mask": "false", "prompt": "legacy prompt"}
+    kwargs = {"headers": H, "data": form}
+    if image_kind == "file":
+        kwargs["files"] = {"image_file": ("frontal.jpg", raw, "image/jpeg")}
+    elif image_kind == "base64":
+        form["image_base64"] = "data:image/jpeg;base64," + base64.b64encode(raw).decode()
+    else:
+        url = "https://legacy.example.test/frontal.jpg"
+        form["image_url"] = url
+        response = httpx.Response(200, content=raw, request=httpx.Request("GET", url))
+        def get_image(actual_url):
+            assert actual_url == url
+            return response
+        monkeypatch.setattr(httpx, "Client", lambda **kw: contextlib.nullcontext(SimpleNamespace(get=get_image)))
+    # No new parameters: old clients keep their original-size overlays.
+    body = client.post(HLGEN52, **kwargs).json()
+    assert body["code"] == 0, body
+    assert {"code", "message", "request_id", "data"}.issubset(body)
+    data = body["data"]
+    assert {"hairline_images", "best_hairline_center_point", "high_hairline_center_point",
+            "low_hairline_center_point", "face_measure"}.issubset(data)
+    assert data["coordinate_space"] == "original_image" and data["framing"] is None
+    assert data["input_image_base64"] == data["original_image_base64"]
+    original = _cv2.imdecode(_np.frombuffer(raw, _np.uint8), _cv2.IMREAD_COLOR)
+    expected = original.shape[:2]
+    assert (data["height"], data["width"]) == expected
+    item = data["hairline_images"][0]
+    assert {"hairline_type", "image_middle_base64", "image_high_base64", "image_low_base64",
+            "grown_image_base64", "order"}.issubset(item)
+    assert item["image_high_base64"] is None and item["image_low_base64"] is None
+    assert data["high_hairline_center_point"] is None and data["low_hairline_center_point"] is None
+    for key in ("image_middle_base64", "grown_image_base64"):
+        image = _cv2.imdecode(_np.frombuffer(base64.b64decode(item[key]), _np.uint8), _cv2.IMREAD_UNCHANGED)
+        assert image.shape[:2] == expected
+    assert len(calls) == 1 and calls[0]["prompt"] == "legacy prompt"
+
+
+@pytest.mark.parametrize("normalize_image", [True, False])
+def test_hairline52_height_compare_uses_same_new_rule(client, monkeypatch, normalize_image):
+    import hairline.comfyui as comfy
+    def unexpected_generation(*args, **kwargs):
+        pytest.fail("Position comparison must not invoke ComfyUI by default")
+    monkeypatch.setattr(comfy, "run", unexpected_generation)
+    with open(fixture("frontal.jpg"), "rb") as photo:
+        response = client.post("/api/v1/debug/hairline-height-compare",
+            files={"image_file": ("frontal.jpg", photo, "image/jpeg")},
+            data={"gender": "male", "hair_style": "1,3", "normalize_image": str(normalize_image).lower()})
+    data = response.json()
+    assert data["code"] == 0, data
+    data = data["data"]
+    assert [r["order"] for r in data["results"]] == [1, 3]
+    assert set(data["guide_landmarks"]) == {"9", "150", "152", "162", "251", "389"}
+    assert data["guide_landmarks"]["9"]["y"] < data["guide_landmarks"]["152"]["y"]
+    expected_size = (data["height"], data["width"])
+    base = _cv2.imdecode(_np.frombuffer(base64.b64decode(data["input_base64"]), _np.uint8), _cv2.IMREAD_COLOR)
+    assert base.shape[:2] == expected_size
+    assert data["coordinate_space"] == ("processed_image" if normalize_image else "original_image")
+    if normalize_image:
+        assert data["framing"]["enabled"] is True
+        assert data["framing"]["axis_x_error_px"] < 1e-8
+        assert data["guide_landmarks"]["9"]["x"] == pytest.approx(data["guide_landmarks"]["152"]["x"], abs=1e-8)
+    else:
+        assert data["framing"] is None
+        original = _cv2.imread(fixture("frontal.jpg"))
+        assert base.shape == original.shape
+    for result in data["results"]:
+        assert result["height_placement"]["target_error_px"] <= 1
+        assert result["grown_image_base64"] is None
+        for key in ("old_overlay_base64", "new_overlay_base64"):
+            assert base64.b64decode(result[key])[:8] == b"\x89PNG\r\n\x1a\n"
+            overlay = _cv2.imdecode(_np.frombuffer(base64.b64decode(result[key]), _np.uint8), _cv2.IMREAD_UNCHANGED)
+            assert overlay.shape[:2] == expected_size
+
+
+def test_hairline52_gen_missing_gender_1004(client):
+    files = {"image_file": ("frontal.jpg", open(fixture("frontal.jpg"), "rb"), "application/octet-stream")}
+    assert client.post(HLGEN52, headers=H, files=files).json()["code"] == 1004
+
+
+def test_hairline52_gen_missing_hairstyle_1007(client):
+    # hair_style 已改为必填（同接口2）：只给 gender、不给 hair_style → 1007
+    files = {"image_file": ("frontal.jpg", open(fixture("frontal.jpg"), "rb"), "application/octet-stream")}
+    assert client.post(HLGEN52, headers=H, files=files, data={"gender": "female"}).json()["code"] == 1007
+
+
+def test_hairline52_gen_female(client, monkeypatch):
+    # mock ComfyUI：验证处理后底图、发际线和生发图使用相同尺寸与坐标。
+    import hairline.comfyui as comfy
+    monkeypatch.setattr(comfy, "run", lambda *a, **k: _PNG_1x1)
+
+    files = {"image_file": ("frontal.jpg", open(fixture("frontal.jpg"), "rb"), "application/octet-stream")}
+    body = client.post(HLGEN52, headers=H, files=files,
+                       data={"gender": "female", "hair_style": "1,3", "normalize_image": "true"}).json()
+    assert body["code"] == 0, body
+    d = body["data"]
+    assert d["coordinate_space"] == "processed_image"
+    assert d["framing"]["enabled"] is True
+    assert d["framing"]["axis_x_error_px"] < 1e-8
+    expected_size = (d["height"], d["width"])
+    for value in (d["input_image_base64"], d["hairline_images"][0]["image_middle_base64"],
+                  d["hairline_images"][0]["grown_image_base64"]):
+        image = _cv2.imdecode(_np.frombuffer(base64.b64decode(value), _np.uint8), _cv2.IMREAD_UNCHANGED)
+        assert image.shape[:2] == expected_size
+    imgs = d["hairline_images"]
+    # 只返回选中发型，order = 发型序号
+    assert [x["order"] for x in imgs] == [1, 3]
+    assert [x["hairline_type"] for x in imgs] == ["ellipse", "heart"]
+    # Only the designed middle curve is generated; deprecated fields stay null.
+    assert base64.b64decode(imgs[0]["image_middle_base64"])[:8] == b"\x89PNG\r\n\x1a\n"
+    assert all(x["image_high_base64"] is None and x["image_low_base64"] is None for x in imgs)
+    assert d["high_hairline_center_point"] is None and d["low_hairline_center_point"] is None
+    rule = d["height_rule"]
+    assert rule["target_upper_height_px"] == pytest.approx(
+        (rule["middle_height_px"] + rule["lower_height_px"]) / 2)
+    for x in imgs:
+        assert x["height_placement"]["new_center"]["y"] == pytest.approx(rule["target_center"]["y"], abs=1)
+    assert base64.b64decode(imgs[0]["grown_image_base64"])[:3] == b"\xff\xd8\xff"
+    assert "image_middle_url" not in imgs[0]
+    c = d["best_hairline_center_point"]
+    assert 0 <= c["x"] < d["width"] and 0 <= c["y"] < d["height"]
+    # face_measure：复用接口1测量数值（容错，失败为 null；本用例正常 → 必须有完整结构）
+    fm = d["face_measure"]
+    assert fm is not None, "face_measure 不应为 null（正常正面照）"
+    assert set(["face_total_height_cm", "four_courts", "seven_eyes",
+                "landmarks", "hairline_source", "head_pose"]).issubset(fm.keys())
+    # 不应含标注图字段（接口52 只要数值，不要画线图）
+    assert "annotated_image_base64" not in fm
+    assert "annotated_image_url" not in fm
+    # 子结构
+    assert set(["top_court_cm", "upper_court_cm", "middle_court_cm",
+                "lower_court_cm", "ratios"]).issubset(fm["four_courts"].keys())
+    assert set(["eye_width_cm", "face_width_cm", "inter_eye_distance_cm",
+                "ratios"]).issubset(fm["seven_eyes"].keys())
+    # 七眼 eye1~eye7 键必须存在（eye1/eye7 在耳朵不可见时可为 null）
+    assert set([f"eye{i}" for i in range(1, 8)]).issubset(fm["seven_eyes"].keys())
+    assert set(["hair_top", "hairline", "brow_center",
+                "nose_bottom", "chin_tip"]).issubset(fm["landmarks"].keys())
+    assert fm["hairline_source"] in ("segmentation", "estimated", "corrected")
+    assert set(["yaw", "pitch", "roll"]).issubset(fm["head_pose"].keys())
+
+
 # 接口4（用户特征）已迁到网关本机实现（直接调豆包），不再在 worker；
 # 其测试随实现一起在网关侧做，worker 这边不再覆盖。

@@ -188,7 +461,7 @@ def test_success_structure(client):
    data = body["data"]
    # 业务字段对齐文档
    assert set(["face_total_height_cm", "four_courts", "seven_eyes",
-                "landmarks", "hairline_source", "head_pose",
+                "landmarks", "hairline_source", "head_pose", "vertical_lines",
                "annotated_image_base64"]).issubset(data.keys())
    assert set(["top_court_cm", "upper_court_cm", "middle_court_cm",
                "lower_court_cm", "ratios"]).issubset(data["four_courts"].keys())
@@ -205,3 +478,40 @@ def test_success_structure(client):
    assert "annotated_image_url" not in data
    png = base64.b64decode(data["annotated_image_base64"])
    assert png[:8] == b"\x89PNG\r\n\x1a\n"
+    _assert_vertical_lines(data, "v1")
+
+
+_INNER_VLINES = [
+    "left_cheek", "left_eye_outer", "left_eye_inner",
+    "right_eye_inner", "right_eye_outer", "right_cheek",
+]
+
+
+def _assert_vertical_lines(data, variant):
+    vl = data["vertical_lines"]
+    assert isinstance(vl["y_top"], int) and isinstance(vl["y_bottom"], int)
+    assert vl["y_top"] < vl["y_bottom"]
+    names = [line["name"] for line in vl["lines"]]
+    xs = [line["x"] for line in vl["lines"]]
+    assert xs == sorted(xs)
+    assert all(isinstance(x, int) for x in xs)
+    if variant == "v6":
+        assert names == _INNER_VLINES
+        assert vl["y_top"] == data["landmarks"]["hairline"]["y"]
+        assert vl["y_bottom"] == data["landmarks"]["chin_tip"]["y"]
+        assert "eye1" not in data["seven_eyes"]
+        assert "eye7" not in data["seven_eyes"]
+        return
+    assert set(_INNER_VLINES).issubset(names)
+    assert names[0] in ("left_head_edge", "left_cheek")
+    assert names[-1] in ("right_head_edge", "right_cheek")
+    by_name = {line["name"]: line["x"] for line in vl["lines"]}
+    assert ("left_head_edge" in by_name) == (data["seven_eyes"]["eye1"] is not None)
+    assert ("right_head_edge" in by_name) == (data["seven_eyes"]["eye7"] is not None)
+
+
+def test_measure_v2_vertical_lines_have_no_head_edges(client):
+    files = {"image_file": ("frontal.jpg", open(fixture("frontal.jpg"), "rb"), "application/octet-stream")}
+    body = client.post("/api/v1/face/measure-v2", headers=H, files=files).json()
+    assert body["code"] == 0, body
+    _assert_vertical_lines(body["data"], "v6")
@@ -0,0 +1,122 @@
+"""Framing must align real pixel coordinates without scaling or invented pixels."""
+import io
+
+import cv2
+import numpy as np
+from PIL import Image
+import pytest
+
+from hairline.framing import calibrate_reference, crop_fits, decode_portrait, normalize_portrait, transform_points
+
+
+PROFILE = {"version": "test", "reference_image_url": "/reference.png",
+           "width_per_face_length": 2.1, "height_per_face_length": 2.8,
+           "brow_x_ratio": .49, "brow_y_ratio": .34}
+
+
+def landmarks(w, h, brow, length=100, angle=0):
+    lm = np.zeros((478, 3), np.float64)
+    down = np.array([np.sin(np.deg2rad(angle)), np.cos(np.deg2rad(angle))])
+    lm[:, :2] = np.array(brow)/[w, h]
+    lm[152, :2] = (np.array(brow)+down*length)/[w, h]
+    lm[:, 2] = .01
+    return lm
+
+
+@pytest.mark.parametrize("angle", [-22, 0, 17])
+def test_native_scale_upright_and_inverse_mapping(angle):
+    photo = np.full((700, 600, 3), 127, np.uint8)
+    lm = landmarks(600, 700, (300, 280), angle=angle)
+    result, moved, metadata = normalize_portrait(photo, lm, PROFILE)
+    h, w = result.shape[:2]
+    assert metadata["standard_crop"] is True
+    assert w/h == .75
+    xy = moved[[9, 152], :2]*[w, h]
+    assert abs(xy[0, 0]-xy[1, 0]) < 1e-9
+    assert xy[1, 1]-xy[0, 1] == pytest.approx(100)
+    assert xy[0, 0]/w == pytest.approx(PROFILE["brow_x_ratio"])
+    assert xy[0, 1]/h == pytest.approx(PROFILE["brow_y_ratio"])
+    matrix = np.array(metadata["source_to_processed"])
+    np.testing.assert_allclose(np.linalg.svd(matrix[:, :2])[1], [1, 1], atol=1e-10)
+    restored = transform_points(moved[:, :2]*[w, h], np.array(metadata["processed_to_source"]))
+    np.testing.assert_allclose(restored, lm[:, :2]*[600, 700], atol=1e-9)
+    assert result.min() == 127  # no black triangles / padded background
+
+
+def test_insufficient_space_keeps_upright_original_size_when_no_rotation_needed():
+    photo = np.full((300, 230, 3), 127, np.uint8)
+    result, moved, metadata = normalize_portrait(photo, landmarks(230, 300, (115, 80), 150), PROFILE)
+    assert metadata["standard_crop"] is False
+    assert metadata["skip_reason"] == "insufficient_space"
+    assert result.shape == photo.shape
+    np.testing.assert_array_equal(result, photo)
+    assert abs(moved[9, 0]-moved[152, 0]) < 1e-10
+
+
+def test_rotated_fallback_has_no_empty_corners():
+    photo = np.full((400, 300, 3), 127, np.uint8)
+    result, moved, metadata = normalize_portrait(photo, landmarks(300, 400, (150, 130), 160, 12), PROFILE)
+    assert metadata["standard_crop"] is False
+    assert result.min() == 127
+    assert metadata["axis_x_error_px"] < 1e-10
+    assert crop_fits(np.array(metadata["source_to_processed"]), (300, 400), result.shape[1::-1])
+
+
+@pytest.mark.parametrize("brow,angle", [((150, 330), 0), ((150, 330), 8), ((150, 70), -8)])
+def test_tall_fallback_is_three_by_four_and_retains_face(brow, angle):
+    photo = np.full((700, 300, 3), 127, np.uint8)
+    lm = landmarks(300, 700, brow, length=200, angle=angle)
+    image, moved, metadata = normalize_portrait(photo, lm, PROFILE)
+    h, w = image.shape[:2]
+    assert metadata["standard_crop"] is False
+    assert metadata["aspect_crop"] is True
+    assert metadata["status"] == "aspect_crop"
+    assert w*4 == h*3
+    xy = moved[:, :2]*[w, h]
+    assert (xy >= -1e-7).all() and (xy <= [w-1+1e-7, h-1+1e-7]).all()
+    assert abs(xy[9, 0]-xy[152, 0]) < 1e-9
+    assert np.linalg.norm(xy[152]-xy[9]) == pytest.approx(200)
+    assert image.min() == 127
+    assert crop_fits(np.array(metadata["source_to_processed"]), (300, 700), (w, h))
+
+
+def test_tall_fallback_does_not_cut_face_that_cannot_fit_three_by_four():
+    photo = np.full((700, 300, 3), 127, np.uint8)
+    image, moved, metadata = normalize_portrait(photo, landmarks(300, 700, (150, 100), length=450), PROFILE)
+    assert metadata["aspect_crop"] is False
+    assert metadata["status"] == "rotation_only"
+    assert image.shape == photo.shape
+
+
+def test_reference_parameters_are_measured_after_straightening():
+    photo = np.full((400, 300, 3), 127, np.uint8)
+    image, moved, profile = calibrate_reference(photo, landmarks(300, 400, (145, 140), angle=8))
+    h, w = image.shape[:2]
+    assert w/h == .75 and w < 300 and h < 400
+    assert profile["brow_9"]["x"] == pytest.approx(profile["chin_152"]["x"], abs=1e-9)
+    assert profile["brow_x_ratio"] == pytest.approx(moved[9, 0])
+    assert profile["brow_y_ratio"] == pytest.approx(moved[9, 1])
+    assert profile["reference_width_per_face_length"] == pytest.approx(w/100)
+    assert profile["reference_height_per_face_length"] == pytest.approx(h/100)
+    assert profile["width_per_face_length"] == 2.5
+    assert profile["height_per_face_length"] == pytest.approx(3/.9)
+    assert image.min() == 127
+
+
+def test_default_crop_makes_face_ten_percent_smaller_than_three_l():
+    photo = np.full((700, 600, 3), 127, np.uint8)
+    result, moved, metadata = normalize_portrait(photo, landmarks(600, 700, (300, 280), length=120))
+    assert metadata["standard_crop"] is True
+    assert result.shape[:2] == (400, 300)
+    assert metadata["face_axis_length_px"] == pytest.approx(120)
+    assert metadata["height_per_face_length"] == pytest.approx(3/.9)
+    assert metadata["width_per_face_length"] == 2.5
+    old_height = 3*120
+    assert old_height/result.shape[0] == pytest.approx(.9)
+
+
+def test_phone_orientation_is_applied_before_landmarks():
+    image = Image.new("RGB", (120, 80), "white")
+    exif = Image.Exif(); exif[274] = 6
+    buf = io.BytesIO(); image.save(buf, format="JPEG", exif=exif)
+    assert decode_portrait(buf.getvalue()).shape[:2] == (120, 80)
@@ -0,0 +1,29 @@
+"""Interface 52 belongs to the GPU worker; the gateway keeps its legacy routes."""
+import asyncio
+import importlib
+
+from starlette.requests import Request
+
+
+def test_legacy_hairline_forwarding(monkeypatch):
+    name, path = "hairline_generate", "/api/v1/hairline/generate"
+    gateway = importlib.import_module("gateway.app")
+    request = Request({"type": "http", "method": "POST", "path": path, "headers": []})
+    calls = []
+
+    async def proxy(actual_request, actual_path):
+        assert actual_request is request
+        calls.append(actual_path)
+        return {"forwarded_to": actual_path}
+
+    monkeypatch.setattr(gateway, "_proxy", proxy)
+    endpoint = getattr(gateway, name)
+    assert next(route for route in gateway.app.routes if getattr(route, "path", None) == path).endpoint is endpoint
+    assert asyncio.run(endpoint(request)) == {"forwarded_to": path}
+    assert calls == [path]
+
+
+def test_interface52_is_not_exposed_by_gateway():
+    gateway = importlib.import_module("gateway.app")
+    assert not any(getattr(route, "path", None) == "/api/v1/hairline/generate-52"
+                   for route in gateway.app.routes)
@@ -0,0 +1,62 @@
+"""Keep the gateway's historical image-field conversion unchanged."""
+import asyncio
+import json
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import httpx
+import pytest
+from starlette.requests import Request
+
+from gateway.forward import rewrite_base64_to_url
+
+
+def test_nullable_image_fields_keep_original_names(tmp_path):
+    value = {"hairline_images": [{"image_high_base64": None, "image_low_base64": None,
+                                   "grown_image_base64": None}],
+             "high_hairline_center_point": None, "face_measure": None}
+    actual = rewrite_base64_to_url(value, "https://example.test", str(tmp_path))
+    assert actual == value
+    assert not list(tmp_path.iterdir())
+
+
+def test_nonempty_images_still_rewrite_alongside_null_fields(tmp_path):
+    import base64
+    import io
+    from PIL import Image
+    buf = io.BytesIO()
+    Image.new("RGBA", (2, 2), (255, 255, 255, 0)).save(buf, format="PNG")
+    value = {"image_middle_base64": base64.b64encode(buf.getvalue()).decode(),
+             "image_high_base64": None}
+    actual = rewrite_base64_to_url(value, "https://example.test", str(tmp_path))
+    assert actual["image_middle_url"].startswith("https://example.test/static/annotations/")
+    assert actual["image_high_base64"] is None
+    assert "image_high_url" not in actual
+    assert len(list(tmp_path.iterdir())) == 1
+
+
+@pytest.mark.parametrize("path,expected_key", [
+    ("/api/v1/hairline/generate", "grown_image_base64"),
+    ("/api/v1/hair/grow", "grown_image_base64"),
+])
+def test_proxy_keeps_original_null_image_fields(tmp_path, monkeypatch, path, expected_key):
+    import gateway.config as config
+    import gateway.forward as forward
+
+    monkeypatch.setattr(config, "get_config", lambda: {
+        "dispatch": {"request_timeout_seconds": 10}, "shared_password": "test-token",
+        "public_base_url": "https://example.test", "static_dir": str(tmp_path),
+    })
+    monkeypatch.setattr(forward, "acquire_worker", AsyncMock(return_value=SimpleNamespace(url="http://worker")))
+    monkeypatch.setattr(forward, "release_worker", AsyncMock())
+    worker_response = httpx.Response(200, json={"code": 0, "data": {"hairline_images": [
+        {"grown_image_base64": None}]}})
+    monkeypatch.setattr(forward, "get_client", lambda: SimpleNamespace(request=AsyncMock(return_value=worker_response)))
+
+    async def receive():
+        return {"type": "http.request", "body": b"", "more_body": False}
+
+    request = Request({"type": "http", "method": "POST", "path": path, "headers": []}, receive)
+    response = asyncio.run(forward.proxy_request(request, path))
+    assert response.status_code == 200
+    assert json.loads(response.body)["data"]["hairline_images"] == [{expected_key: None}]
@@ -27,5 +27,9 @@ def test_mp_to_obj_anchor_mapping():

 def test_texture_map():
    m = get_texture_map()
-    assert [k for k, _ in m["female"]] == ["ellipse", "flower", "heart", "straight", "wave"]
-    assert [k for k, _ in m["male"]] == ["ellipse", "inverse_arc", "m", "straight"]
+    assert [k for k, _ in m["female"]] == [
+        "ellipse", "flower", "heart", "straight", "wave", "bigflower", "clasicalflower",
+    ]
+    assert [k for k, _ in m["male"]] == [
+        "ellipse", "inverse_arc", "m", "straight", "heart", "Softpetal", "wave",
+    ]
@@ -0,0 +1,82 @@
+"""Template placement must follow face pose, not unequal left/right bangs."""
+import numpy as np
+import pytest
+
+from hairline import constants as C
+from hairline.hairline_2d import (
+    align_hairline_for_template,
+    clamp_hairline_to_silhouette,
+)
+
+
+def _scene(angle=0.0, shape=(600, 900)):
+    h, w = shape
+    scale = np.array([w, h])
+    lm = np.zeros((468, 3), np.float32)
+    xy = np.full((468, 2), [450.0, 300.0])
+    for i, (left, right) in enumerate(C.MP_TOP_ANCHOR_PAIRS):
+        xy[left] = [300.0 + i * 12, 210.0 - i * 8]
+        xy[right] = [600.0 - i * 12, 210.0 - i * 8]
+    xy[10] = [450, 150]
+    xy[152] = [450, 450]
+    theta = np.deg2rad(angle)
+    rotation = np.array([[np.cos(theta), -np.sin(theta)],
+                         [np.sin(theta), np.cos(theta)]])
+    xy = (xy - [450, 300]) @ rotation.T + [450, 300]
+    lm[:, :2] = xy / scale
+    up = rotation @ [0.0, -1.0]
+    anchors = lm[C.MP_TOP_ANCHORS, :2] * scale
+    distances = np.full(C.N_ANCHORS, 50.0)
+    indices = {k: i for i, k in enumerate(C.MP_TOP_ANCHORS)}
+    distances[indices[67]] = 80.0
+    distances[indices[297]] = 20.0
+    hairline = ((anchors + distances[:, None] * up) / scale).astype(np.float32)
+    return lm, hairline, np.ones(C.N_ANCHORS, bool), anchors, up
+
+
+@pytest.mark.parametrize("angle,shape", [(0, (600, 900)), (23, (600, 900)),
+                                         (-17, (1200, 900))])
+def test_uneven_bangs_do_not_shear_template(angle, shape):
+    lm, raw, valid, anchors, up = _scene(angle, shape)
+    raw_copy, lm_copy = raw.copy(), lm.copy()
+    aligned = align_hairline_for_template(lm, raw, valid, shape)
+    displacement = aligned * shape[::-1] - anchors
+    # Every anchor extends 50 px along the rotated face axis, even when
+    # one side's hair boundary is 60 px higher than its counterpart.
+    np.testing.assert_allclose(displacement, np.tile(50 * up, (C.N_ANCHORS, 1)), atol=1e-4)
+    np.testing.assert_array_equal(raw, raw_copy)
+    np.testing.assert_array_equal(lm, lm_copy)
+
+
+def test_anatomical_pairs_are_not_reversed_anchor_order():
+    shape = (600, 900)
+    lm, raw, valid, anchors, up = _scene(shape=shape)
+    # Endpoints of MP_TOP_ANCHORS are not mirror-ordered.
+    distances = {127: 10, 356: 30, 234: 60, 454: 100}
+    for mp_idx, distance in distances.items():
+        i = C.MP_TOP_ANCHORS.index(mp_idx)
+        raw[i] = (anchors[i] + distance * up) / shape[::-1]
+    aligned = align_hairline_for_template(lm, raw, valid, shape)
+    actual = (aligned * shape[::-1] - anchors) @ up
+    for mp_idx, expected in [(127, 20), (356, 20), (234, 80), (454, 80)]:
+        assert actual[C.MP_TOP_ANCHORS.index(mp_idx)] == pytest.approx(expected, abs=1e-4)
+
+
+def test_missing_ray_hit_does_not_pull_valid_partner():
+    shape = (600, 900)
+    lm, raw, valid, _anchors, _up = _scene(shape=shape)
+    left, right = C.MP_TOP_ANCHORS.index(67), C.MP_TOP_ANCHORS.index(297)
+    valid[right] = False
+    raw[right, 1] = -0.5  # An extrapolated fallback outside the head.
+    aligned = align_hairline_for_template(lm, raw, valid, shape)
+    np.testing.assert_allclose(aligned[[left, right]], raw[[left, right]], atol=1e-7)
+
+
+def test_template_alignment_still_allows_silhouette_clamping():
+    shape = (600, 900)
+    lm, raw, valid, _anchors, _up = _scene(shape=shape)
+    aligned = align_hairline_for_template(lm, raw, valid, shape)
+    parse_map = np.zeros(shape, np.int32)
+    parse_map[140:450, 250:650] = C.PARSE_SKIN
+    clamped = clamp_hairline_to_silhouette(aligned, parse_map)
+    assert np.min(clamped[:, 1] * shape[0]) >= 142 - 1e-4
@@ -0,0 +1,91 @@
+"""Court-based design placement, including preview/inpaint consistency."""
+import cv2
+import numpy as np
+import pytest
+
+from hairline.height_rule import (
+    height_target, curve_center, place_curve, inpaint_from_curve,
+)
+
+
+def _landmarks(width=320, height=400):
+    lm = np.zeros((468, 3), np.float32)
+    for index, xy in {9: (160, 180), 94: (160, 260), 152: (160, 320),
+                      10: (160, 140), 21: (70, 170), 251: (250, 170)}.items():
+        lm[index, :2] = np.array(xy) / [width, height]
+    return lm
+
+
+def test_target_uses_mean_courts_without_old_estimation_ratio():
+    rule = height_target(_landmarks(), 320, 400)
+    assert rule["middle_height_px"] == pytest.approx(80, abs=1e-4)
+    assert rule["lower_height_px"] == pytest.approx(60, abs=1e-4)
+    assert rule["target_upper_height_px"] == pytest.approx(70, abs=1e-4)
+    assert rule["target_center"]["y"] == pytest.approx(110, abs=1e-4)
+
+
+def test_reference_uses_unrounded_measurement_and_scale():
+    ref = {"brow_center": (161.4, 181.3), "nose_bottom": (160.0, 261.9),
+           "chin_tip": (160.0, 321.5), "px_per_cm": 20}
+    rule = height_target(_landmarks(), 320, 400, ref)
+    assert rule["target_upper_height_px"] == pytest.approx(70.1)
+    assert rule["target_center"]["y"] == pytest.approx(111.2)
+    assert rule["target_upper_height_cm"] == pytest.approx(3.505)
+    assert rule["measurement_source"] == "face_measure"
+
+
+@pytest.mark.parametrize("scale", [1, 2])
+@pytest.mark.parametrize("center_y", [90, 125, 150])
+def test_all_shapes_align_to_same_height_at_multiple_scales(scale, center_y):
+    h, w = 400 * scale, 320 * scale
+    lm = _landmarks()
+    overlay = np.zeros((h, w, 4), np.float32)
+    # A non-horizontal design, preserving a relative 20px arc after movement.
+    pts = np.array([[70, center_y + 20], [160, center_y], [250, center_y + 20]]) * scale
+    cv2.polylines(overlay, [pts.astype(np.int32)], False, (255, 255, 255, 255), 3 * scale)
+    original = overlay.copy()
+    target = height_target(lm, w, h)["target_center"]
+    moved, placement = place_curve(overlay, target, lm)
+    assert abs(curve_center(moved, target["x"])["y"] - target["y"]) < .2
+    assert placement["target_error_px"] < .2
+    # Shape's relative heights stay constant after translation.
+    before_delta = curve_center(original, 100 * scale)["y"] - curve_center(original, 160 * scale)["y"]
+    after_delta = curve_center(moved, 100 * scale)["y"] - curve_center(moved, 160 * scale)["y"]
+    assert after_delta == pytest.approx(before_delta, abs=.1)
+    np.testing.assert_array_equal(overlay, original)
+
+
+def test_rolled_face_preserves_curve_angle():
+    lm = _landmarks()
+    lm[10, 0] -= .08
+    overlay = np.zeros((400, 320, 4), np.float32)
+    cv2.line(overlay, (60, 95), (260, 135), (255, 255, 255, 255), 3)
+    moved, _ = place_curve(overlay, {"x": 160, "y": 150}, lm)
+    original_slope = curve_center(overlay, 200)["y"] - curve_center(overlay, 120)["y"]
+    moved_slope = curve_center(moved, 200)["y"] - curve_center(moved, 120)["y"]
+    assert moved_slope == pytest.approx(original_slope, abs=.2)
+
+
+def test_inpaint_mask_follows_new_line_not_old_black_template():
+    lm = _landmarks()
+    photo = np.full((400, 320, 3), 200, np.uint8)
+    parsing = np.ones((400, 320), np.int32)
+    overlay = np.zeros((400, 320, 4), np.float32)
+    cv2.line(overlay, (20, 140), (300, 140), (255, 255, 255, 255), 3)
+    moved, _ = place_curve(overlay, {"x": 160, "y": 100}, lm)
+    marked, mask = inpaint_from_curve(photo, lm, parsing, moved)
+    assert marked[100, 160, 0] == 0
+    assert marked[140, 160, 0] == 200
+    assert mask[90, 160] > 240
+    assert mask[130, 160] < 10
+
+
+def test_out_of_frame_target_is_reported_instead_of_silent_clamping():
+    ref = {"brow_center": (160, 50), "nose_bottom": (160, 200), "chin_tip": (160, 350)}
+    with pytest.raises(ValueError, match="超出图片"):
+        height_target(_landmarks(), 320, 400, ref)
+
+
+def test_missing_midline_curve_is_reported():
+    with pytest.raises(ValueError, match="有效曲线"):
+        place_curve(np.zeros((400, 320, 4), np.float32), {"x": 160, "y": 100}, _landmarks())
@@ -0,0 +1,101 @@
+import json
+from ipaddress import ip_address
+
+import pytest
+from fastapi.testclient import TestClient
+from starlette.requests import Request
+
+import ip_access as access
+
+
+@pytest.fixture
+def policy(tmp_path, monkeypatch):
+    path = tmp_path / "access.json"
+    path.write_text(json.dumps({"allowed_ips": []}))
+    monkeypatch.setattr(access, "CONFIG_PATH", path)
+    monkeypatch.setattr(access, "local_addresses", lambda: {ip_address("192.168.10.2"), ip_address("::1")})
+    return path
+
+
+def request(host, headers=None, port=50000):
+    return Request({"type": "http", "path": "/api/v1/hairline/generate-52", "scheme": "http",
+                    "client": (host, port), "headers": [(k.encode(), v.encode()) for k, v in (headers or {}).items()]})
+
+
+@pytest.mark.parametrize("host", ["127.0.0.1", "::1", "::ffff:127.0.0.1", "192.168.10.2"])
+def test_only_actual_local_addresses_allowed_by_default(policy, host):
+    assert access.check_access(request(host))[0]
+
+
+@pytest.mark.parametrize("host", ["203.0.113.9", "192.168.10.3", "0.0.0.0", "invalid"])
+def test_other_addresses_denied(policy, host):
+    assert not access.check_access(request(host))[0]
+
+
+def test_whitelist_ipv4_ipv6_cidr_and_hot_reload(policy):
+    policy.write_text(json.dumps({"allowed_ips": ["203.0.113.9", "198.51.100.0/24", "2001:db8::/64"]}))
+    for host in ["203.0.113.9", "::ffff:203.0.113.9", "198.51.100.12", "2001:db8::12"]:
+        assert access.check_access(request(host))[0]
+    policy.write_text('{"allowed_ips": []}')
+    assert not access.check_access(request("203.0.113.9"))[0]
+
+
+def test_configured_public_whitelist_is_exact(policy):
+    policy.write_text('{"allowed_ips": ["49.232.51.171"]}')
+    assert access.check_access(request("49.232.51.171"))[0]
+    assert not access.check_access(request("49.232.51.172"))[0]
+
+
+@pytest.mark.parametrize("header", ["x-forwarded-for", "x-real-ip", "forwarded"])
+def test_forwarding_headers_cannot_override_socket_peer(policy, header):
+    assert not access.check_access(request("203.0.113.9", {header: "127.0.0.1"}))[0]
+    assert access.check_access(request("127.0.0.1", {header: "203.0.113.9"}))[0]
+
+
+def test_gateway_is_allowed_independently_of_end_user_ip(policy):
+    policy.write_text('{"allowed_ips": ["49.232.51.171"]}')
+    assert access.check_access(request("49.232.51.171", {"x-forwarded-for": "203.0.113.9"}))[0]
+    assert not access.check_access(request("203.0.113.9", {"x-forwarded-for": "49.232.51.171"}))[0]
+
+
+def test_rewritten_peer_or_invalid_policy_fail_closed(policy):
+    assert not access.check_access(request("127.0.0.1", port=0))[0]
+    policy.write_text('{"allowed_ips": ["not-an-ip"]}')
+    assert not access.check_access(request("127.0.0.1"))[0]
+    policy.write_text('invalid JSON')
+    assert not access.check_access(request("127.0.0.1"))[0]
+
+
+def test_worker_guard_covers_all_interfaces_before_auth(policy, monkeypatch):
+    import app as worker
+    monkeypatch.setattr(worker, "ACCEPT_PASSWORDS", ["access-test"])
+    headers = {"X-Internal-Token": "access-test"}
+    with TestClient(worker.app, client=("203.0.113.9", 50000)) as client:
+        assert client.post('/api/v1/hairline/generate-52').status_code == 403
+        assert client.post('/api/v1/hairline/generate-52/', headers=headers).status_code == 403
+        assert client.post('/api/v1/debug/hairline-height-compare').status_code == 403
+        assert client.post('/api/v1/hairline/generate', headers=headers).status_code == 403
+        assert client.post('/api/v1/face/measure', headers=headers).status_code == 403
+        assert client.post('/api/v1/redraw', headers=headers).status_code == 403
+        assert client.get('/health').status_code == 403
+        assert client.get('/openapi.json').status_code == 403
+        policy.write_text('{"allowed_ips": ["203.0.113.9"]}')
+        assert client.post('/api/v1/hairline/generate-52', headers=headers).json()['code'] == 1004
+        assert client.post('/api/v1/hairline/generate-52').status_code == 401
+
+
+def test_gateway_does_not_apply_worker_ip_policy(policy, monkeypatch):
+    import importlib
+    gateway = importlib.import_module('gateway.app')
+    calls = []
+    async def proxy(request, path):
+        calls.append(path)
+        return {'code': 0}
+    monkeypatch.setattr(gateway, '_proxy', proxy)
+    # Deliberately skip lifespan: no workers or configuration are needed.
+    client = TestClient(gateway.app, client=('203.0.113.9', 50000))
+    assert client.post('/api/v1/hairline/generate-52').status_code == 404
+    assert client.post('/api/v1/hairline/generate').json()['code'] == 0
+    assert client.post('/api/v1/hair/grow').json()['code'] == 0
+    assert client.get('/static/test_interface52.html').status_code == 200
+    assert calls == ['/api/v1/hairline/generate', '/api/v1/hair/grow']
@@ -0,0 +1,83 @@
+"""竖线坐标：与标注图同一套几何，接口1/6 的返回字段由此而来。"""
+import numpy as np
+
+from face_analysis.measure import MeasureResult
+from face_analysis.annotation import vertical_line_geometry
+
+
+def _result():
+    """已知像素坐标的测量结果。顶庭 2cm，不会触发发际线修正。"""
+    vertical = {
+        "hair_top": (500.0, 100.0),
+        "hairline": (500.0, 200.2),
+        "brow_center": (500.0, 350.0),
+        "nose_bottom": (500.0, 500.0),
+        "chin_tip": (500.0, 700.4),
+        "top_court_px": 100.0,
+        "upper_court_px": 150.0,
+        "middle_court_px": 150.0,
+        "lower_court_px": 200.0,
+    }
+    eyes = {
+        "eye_width_px": 80.0,
+        "face_width_px": 400.0,
+        "inter_eye_distance_px": 90.0,
+        "points": {
+            "left_cheek": (300.4, 400.0),
+            "left_outer": (360.6, 400.0),
+            "left_inner": (440.2, 400.0),
+            "right_inner": (530.8, 400.0),
+            "right_outer": (610.4, 400.0),
+            "right_cheek": (700.7, 400.0),
+        },
+    }
+    return MeasureResult(vertical, eyes, 50.0, "estimated", None)
+
+
+_INNER = [
+    "left_cheek", "left_eye_outer", "left_eye_inner",
+    "right_eye_inner", "right_eye_outer", "right_cheek",
+]
+
+
+def test_v1_inner_lines_and_extended_span():
+    geom = vertical_line_geometry(_result(), 1000, 800, variant="v1")
+    # 短边 800，伸出 max(6, round(800*0.03)) = 24
+    assert geom["y_top"] == 76
+    assert geom["y_bottom"] == 724
+    assert [line["name"] for line in geom["lines"]] == _INNER
+    assert [line["x"] for line in geom["lines"]] == [300, 361, 440, 531, 610, 701]
+
+
+def test_v6_span_is_hairline_to_chin_without_head_edges():
+    ear = np.zeros((800, 1000), dtype=np.uint8)
+    ear[120:680, 250] = 255
+    ear[120:680, 760] = 255
+    geom = vertical_line_geometry(
+        _result(), 1000, 800, ear_mask=ear, hair_mask=None, variant="v6")
+    assert geom["y_top"] == 200
+    assert geom["y_bottom"] == 700
+    assert [line["name"] for line in geom["lines"]] == _INNER
+
+
+def test_v1_includes_ear_edges_left_to_right():
+    ear = np.zeros((800, 1000), dtype=np.uint8)
+    ear[120:680, 250] = 255
+    ear[120:680, 760] = 255
+    geom = vertical_line_geometry(
+        _result(), 1000, 800, ear_mask=ear, hair_mask=None, variant="v1")
+    names = [line["name"] for line in geom["lines"]]
+    xs = [line["x"] for line in geom["lines"]]
+    assert names == ["left_head_edge", *_INNER, "right_head_edge"]
+    assert xs[0] == 250 and xs[-1] == 760
+    assert xs == sorted(xs)
+
+
+def test_missing_ear_omits_that_edge():
+    ear = np.zeros((800, 1000), dtype=np.uint8)
+    ear[120:680, 250] = 255
+    geom = vertical_line_geometry(
+        _result(), 1000, 800, ear_mask=ear, hair_mask=None, variant="v1")
+    names = [line["name"] for line in geom["lines"]]
+    assert names[0] == "left_head_edge"
+    assert "right_head_edge" not in names
@@ -1,5 +1,11 @@
 {
-  "_comment": "复制为 worker_config.json（不入 git）。accept_passwords：worker 内网鉴权密码列表，网关带 X-Internal-Token: <其中之一>（也可用环境变量 WORKER_ACCEPT_PASSWORDS 覆盖）。comfyui_password：本机 ComfyUI(8188) Basic Auth 密码（user 默认 admin；也可放 password.txt 或环境变量 COMFYUI_PASSWORD）。注：接口4 火山方舟 ark_api_key 已迁到网关配置。",
+  "_comment": "复制为 worker_config.json（不入 git）。accept_passwords：worker 内网鉴权密码列表，网关带 X-Internal-Token: <其中之一>（也可用环境变量 WORKER_ACCEPT_PASSWORDS 覆盖）。comfyui_password：本机 ComfyUI(8188) Basic Auth 密码（user 默认 admin；也可放 password.txt 或环境变量 COMFYUI_PASSWORD）。health：深健康检查（/health 磁盘水位+下游依赖 TCP 探测，均可用环境变量 HEALTH_MIN_FREE_GB / HEALTH_CHECK_DEPS / HEALTH_DEP_COMFYUI / HEALTH_DEPS_CACHE_SECONDS 覆盖）。注：接口4 火山方舟 ark_api_key 已迁到网关配置。",
  "accept_passwords": ["change-me-to-a-strong-secret"],
-  "comfyui_password": "your-comfyui-basic-auth-password"
+  "comfyui_password": "your-comfyui-basic-auth-password",
+  "health": {
+    "min_free_gb": 5,
+    "check_deps": true,
+    "dep_comfyui": "127.0.0.1:8188",
+    "deps_cache_seconds": 5
+  }
 }
Author	SHA1	Message	Date
xslandCursor	ccb9c142c4	docs: 接入页补充接口1/6 的竖线坐标 Co-authored-by: Cursor <cursoragent@cursor.com>	2026-11-05 22:29:59 +08:00
xslandCursor	b64313dc73	feat(测量): 接口1/6 返回与标注图一致的竖线坐标前端需要按标注图重绘竖线，坐标与画线共用同一套几何。 Co-authored-by: Cursor <cursoragent@cursor.com>	2026-11-05 22:18:12 +08:00
xsl	1c51f3eaab	fix: 将52号接口范围限定为GPU worker 撤回52号网关路由、图片字段改名和mock改动，gateway目录恢复a3fddbb。worker保持base64和null响应，更新接口说明和回归测试。验证：worker API、IP限制与网关兼容性共54项测试通过。	2026-11-04 10:45:16 +08:00
xsl	5b2b73f5d5	feat: 新增52号发际线接口并限制GPU worker来源IP 保留5号历史三档实现，新增中下庭高度定位、扶正裁切、矩形叠加与30案例对比页。 worker仅允许本机和配置白名单访问；网关不增加IP限制。空图片URL字段改写限定为52号，保持旧接口兼容。同步接口、裁切与部署文档。验证：113项测试通过；1项历史数值基线失败，在未修改的a3fddbb版本复现。浏览器验证5/52及30案例对比页通过。	2026-11-04 10:20:29 +08:00
xsl	a3fddbbc34	fix(发际线): 修正不对称刘海导致的模板叠加倾斜按 MediaPipe 左右对应额头点协调模板延伸距离，保留人脸姿态与头部轮廓钳制，统一预览白线和生发遮罩的几何。补充对齐回归测试并更新贴图顺序测试及实现说明。验证：13 项相关测试通过；两张用户图片完成前后对比，新图男性 7 型接口与生发图均成功返回。	2026-11-01 15:14:29 +08:00
xsl	8554850d55	docs: 网关摘除隔离期实施方案（供网关机器参考实现，暂不在本仓库改 gateway 代码）复盘 9·19 磁盘满事故网关未切换的三层原因；给出 main 分支 gateway/pool.py 的隔离期改动说明（WorkerState.offline_until + 被动摘除冷却 60s + 探测复活守卫 + 删死代码）、配置项、单测建议、部署/验证/回滚步骤。worker 侧深检（方案 A）已在本分支 `14562d4` 上线。	2026-09-20 22:07:48 +08:00
xsl	6b098c0e15	feat(health): /health 深检——磁盘水位 + 下游依赖 TCP 探测，异常返回 503 degraded 背景: 9·19 磁盘满事故中业务已瘫但 /health 仍 200（只看模型就绪标志），网关持续把流量打到本机。深检让 worker 在磁盘不足(min_free_gb, 默认5G)或 ComfyUI/change_hair TCP 不通时主动报 503，网关探测两次失败即摘除。 - 配置: worker_config.json "health" 段，env HEALTH_MIN_FREE_GB/ HEALTH_CHECK_DEPS/HEALTH_DEP_COMFYUI/HEALTH_DEPS_CACHE_SECONDS 可覆盖 - 依赖探测 0.8s 超时并发执行，结果缓存 5s；深检自身异常按就绪返回 - 测试: 深检各分支用例 + 旧用例改为 mock 深检（不依赖本机服务）	2026-09-20 22:07:48 +08:00