sophhub 0.4.19 → 0.4.21

package/skills/sophnet-id-photo/src/scripts/id_photo_compliance.py

@@ -1,215 +1,215 @@
-"""
-证件照输入合规预审：调用视觉模型 Qwen3-VL-235B-A22B-Instruct，判定图片是否适合做
-「换底 + 轻量校正/裁剪」类预处理。本模块归属 sophnet-id-photo，不依赖其它 skill。
-"""
-
-import argparse
-import base64
-import json
-import os
-import re
-import sys
-import urllib.error
-import urllib.request
-
-import sophnet_tools
-
-API_URL = "https://www.sophnet.com/api/open-apis/v1/chat/completions"
-MODEL_ID_PHOTO_COMPLIANCE = "Qwen3-VL-235B-A22B-Instruct"
-
-ID_PHOTO_STANDARD_BRIEF_CROPPED = (
-    "当前画面为**算法裁切后的单人头肩区域**（默认仅保留一名主体），**勿再以合照或非单人**为由拒审。"
-    "**主要拦截**："
-    "（1）**口罩、墨镜**：遮挡口鼻的口罩，或遮挡眼部的墨镜（**透明光学眼镜可保留**）；"
-    "（2）**人脸被遮挡**：头发、手、物体等遮挡面部或五官，影响辨认；"
-    "（3）**双耳未在可见范围内露出**：双耳被头发、配饰或姿态完全挡住。"
-    "其余宜通过：真人、彩色、非卡通/截图，脸部大致可辨。**不要**因背景杂物、衣着拒审。"
-)
-
-ID_PHOTO_STANDARD_BRIEF = (
-    "预审**不限定场景**（杂乱背景可由后续换纯色底处理），**主要拦截**以下硬伤："
-    "（1）**合照或非单人**：多人合影、或多名清晰人物并列导致无法认定唯一要处理的主体；"
-    "（2）**口罩、墨镜**：遮挡口鼻的口罩，或遮挡眼部的墨镜（**透明光学眼镜可保留**）；"
-    "（3）**人脸被遮挡**：头发、手、物体等遮挡面部或五官，影响辨认；"
-    "（4）**双耳未在可见范围内露出**：双耳被头发、配饰或姿态完全挡住，不符合证件照常见露耳要求。"
-    "其余情况宜判定为可通过：真人、彩色、非卡通/截图，脸部大致可辨。"
-    "若人物在画面中偏小或远景/全身取景，只要上述硬伤不存在、双耳与人脸仍可据图判断，**仍宜通过**"
-    "（后续会先做人脸检测与头肩裁切再审）。"
-    "除所列硬伤外，**不要**因背景类型、生活/户外环境、衣着或杂物等拒审。"
-)
-
-
-def _is_likely_base64(raw: str) -> bool:
-    text = raw.strip()
-    if not text or any(ch.isspace() for ch in text):
-        return False
-    try:
-        base64.b64decode(text, validate=True)
-        return True
-    except Exception:
-        return False
-
-
-def normalize_image_input(image_input: str, upload_timeout: int = 30) -> str:
-    text = image_input.strip()
-    if not text:
-        raise ValueError("图片输入不能为空")
-
-    if text.startswith(("http://", "https://")):
-        return text
-
-    if text.startswith("data:image/") and ";base64," in text:
-        return text
-
-    if os.path.exists(text):
-        return sophnet_tools.upload_oss(text, upload_timeout)
-
-    if _is_likely_base64(text):
-        return text
-
-    raise ValueError("无法识别图片输入：请传入 URL、base64、data URL 或本地文件路径")
-
-
-def _post_chat_completions(payload: dict) -> dict:
-    api_key = sophnet_tools.get_api_key()
-    if not api_key:
-        raise RuntimeError("未配置 API Key，无法进行证件照合规预审")
-    body = json.dumps(payload).encode("utf-8")
-    request = urllib.request.Request(
-        url=API_URL,
-        data=body,
-        method="POST",
-        headers={
-            "Content-Type": "application/json",
-            "Accept": "application/json",
-            "Authorization": f"Bearer {api_key}",
-        },
-    )
-    try:
-        with urllib.request.urlopen(request, timeout=180) as response:
-            raw = response.read().decode("utf-8")
-            return json.loads(raw)
-    except urllib.error.HTTPError as exc:
-        detail = exc.read().decode("utf-8", errors="ignore")
-        raise RuntimeError(f"HTTPError: {exc.code} {exc.reason} | {detail}") from exc
-    except urllib.error.URLError as exc:
-        raise RuntimeError(f"URLError: {exc.reason}") from exc
-
-
-def _extract_message_text(data: dict) -> str:
-    try:
-        return data["choices"][0]["message"]["content"]
-    except (KeyError, IndexError, TypeError):
-        return json.dumps(data, ensure_ascii=False, indent=2)
-
-
-def _strip_json_fence(text: str) -> str:
-    t = text.strip()
-    m = re.match(r"^```(?:json)?\s*([\s\S]*?)\s*```\s*$", t, re.IGNORECASE)
-    if m:
-        return m.group(1).strip()
-    return t
-
-
-def check_id_photo_compliance(image_input: str, *, image_scope: str = "full") -> dict:
-    """
-    返回 dict：compliant, summary, violations, standard_brief, confidence, model；含 raw 供调试。
-    image_scope: \"full\" 整张原图预审；\"cropped\" 裁切后头肩区预审（规范与 system 提示更短，不再强调合照）。
-    """
-    image_url = normalize_image_input(image_input)
-    is_cropped = image_scope == "cropped"
-    standard = ID_PHOTO_STANDARD_BRIEF_CROPPED if is_cropped else ID_PHOTO_STANDARD_BRIEF
-
-    if is_cropped:
-        system = (
-            "你是证件照预处理预审助手。图为**已裁切的单人头肩候选区**，勿编造细节。"
-            "**主要拦截**：佩戴口罩、佩戴墨镜、人脸被遮挡、双耳未露出；**不要**以合照、场景、背景为由拒审。"
-            "violations 措辞优先：「请勿佩戴口罩」「请勿佩戴墨镜」「人脸被遮挡」「双耳未露出」。"
-            "口罩与墨镜须分两条 violation。无硬伤时 compliant=true、violations=[]。"
-            "输出且仅输出 JSON：compliant（布尔）、summary（一至两句中文）、violations（字符串数组）、"
-            "confidence（high/medium/low）。"
-        )
-    else:
-        system = (
-            "你是证件照预处理预审助手。根据用户给出的「规范要点」判断 compliant，不编造图中没有的细节。"
-            "**主要拦截**：合照或无法认定单人主体、佩戴口罩、佩戴墨镜、人脸被遮挡、双耳未露出。"
-            "除此之外宜判定通过；**不要**针对具体场景（如咖啡馆、街道、室内陈设等）或手持物品类别做特化拒审。"
-            "violations 为若干条**简短中文**，命中硬伤时**逐条列出**；"
-            "措辞优先且尽量分列：「疑似合照，无法认定单一人物主体」「请勿佩戴口罩」「请勿佩戴墨镜」「人脸被遮挡」「双耳未露出」。"
-            "若同时存在口罩与墨镜，须分成两条 violation，勿合并为一条。"
-            "无硬伤时 compliant=true 且 violations=[]。"
-            "输出且仅输出一个 JSON，不要 Markdown："
-            '键 compliant（布尔）、summary（一至两句中文）、violations（字符串数组）、'
-            "confidence（high/medium/low）。"
-        )
-    user = (
-        "【规范要点】\n"
-        + standard
-        + "\n\n请审视所附图片，只输出 JSON。"
-    )
-    payload = {
-        "messages": [
-            {"role": "system", "content": system},
-            {
-                "role": "user",
-                "content": [
-                    {"type": "text", "text": user},
-                    {"type": "image_url", "image_url": {"url": image_url}},
-                ],
-            },
-        ],
-        "model": MODEL_ID_PHOTO_COMPLIANCE,
-        "stream": False,
-    }
-    data = _post_chat_completions(payload)
-    raw = _extract_message_text(data)
-    try:
-        obj = json.loads(_strip_json_fence(raw))
-    except json.JSONDecodeError as e:
-        raise RuntimeError(f"合规判定返回非 JSON：{raw[:500]}...") from e
-
-    compliant = bool(obj.get("compliant"))
-    violations = obj.get("violations") or []
-    if not isinstance(violations, list):
-        violations = [str(violations)]
-    summary = str(obj.get("summary") or ("符合规范" if compliant else "不符合规范"))
-    return {
-        "compliant": compliant,
-        "summary": summary,
-        "violations": [str(v) for v in violations],
-        "standard_brief": standard,
-        "confidence": str(obj.get("confidence") or "unknown"),
-        "model": MODEL_ID_PHOTO_COMPLIANCE,
-        "raw": raw,
-    }
-
-
-def build_parser():
-    p = argparse.ArgumentParser(description="证件照输入合规预审（视觉模型，本 skill 自包含）")
-    p.add_argument(
-        "--scope",
-        choices=("full", "cropped"),
-        default="full",
-        help="full=整张原图「规范要点」；cropped=头肩裁切后精简要点（不再强调合照）",
-    )
-    p.add_argument(
-        "image",
-        help="图片：http(s) URL、data URL、base64 或本地路径（本地会先上传）",
-    )
-    return p
-
-
-def main():
-    args = build_parser().parse_args()
-    try:
-        report = check_id_photo_compliance(args.image, image_scope=args.scope)
-    except Exception as e:
-        print(str(e), file=sys.stderr)
-        sys.exit(1)
-    out = {k: v for k, v in report.items() if k != "raw"}
-    print(json.dumps(out, ensure_ascii=False))
-    sys.exit(0 if report["compliant"] else 2)
-
-
-if __name__ == "__main__":
-    main()
+"""
+证件照输入合规预审：调用视觉模型 Qwen3-VL-235B-A22B-Instruct，判定图片是否适合做
+「换底 + 轻量校正/裁剪」类预处理。本模块归属 sophnet-id-photo，不依赖其它 skill。
+"""
+
+import argparse
+import base64
+import json
+import os
+import re
+import sys
+import urllib.error
+import urllib.request
+
+import sophnet_tools
+
+API_URL = "https://www.sophnet.com/api/open-apis/v1/chat/completions"
+MODEL_ID_PHOTO_COMPLIANCE = "Qwen3-VL-235B-A22B-Instruct"
+
+ID_PHOTO_STANDARD_BRIEF_CROPPED = (
+    "当前画面为**算法裁切后的单人头肩区域**（默认仅保留一名主体），**勿再以合照或非单人**为由拒审。"
+    "**主要拦截**："
+    "（1）**口罩、墨镜**：遮挡口鼻的口罩，或遮挡眼部的墨镜（**透明光学眼镜可保留**）；"
+    "（2）**人脸被遮挡**：头发、手、物体等遮挡面部或五官，影响辨认；"
+    "（3）**双耳未在可见范围内露出**：双耳被头发、配饰或姿态完全挡住。"
+    "其余宜通过：真人、彩色、非卡通/截图，脸部大致可辨。**不要**因背景杂物、衣着拒审。"
+)
+
+ID_PHOTO_STANDARD_BRIEF = (
+    "预审**不限定场景**（杂乱背景可由后续换纯色底处理），**主要拦截**以下硬伤："
+    "（1）**合照或非单人**：多人合影、或多名清晰人物并列导致无法认定唯一要处理的主体；"
+    "（2）**口罩、墨镜**：遮挡口鼻的口罩，或遮挡眼部的墨镜（**透明光学眼镜可保留**）；"
+    "（3）**人脸被遮挡**：头发、手、物体等遮挡面部或五官，影响辨认；"
+    "（4）**双耳未在可见范围内露出**：双耳被头发、配饰或姿态完全挡住，不符合证件照常见露耳要求。"
+    "其余情况宜判定为可通过：真人、彩色、非卡通/截图，脸部大致可辨。"
+    "若人物在画面中偏小或远景/全身取景，只要上述硬伤不存在、双耳与人脸仍可据图判断，**仍宜通过**"
+    "（后续会先做人脸检测与头肩裁切再审）。"
+    "除所列硬伤外，**不要**因背景类型、生活/户外环境、衣着或杂物等拒审。"
+)
+
+
+def _is_likely_base64(raw: str) -> bool:
+    text = raw.strip()
+    if not text or any(ch.isspace() for ch in text):
+        return False
+    try:
+        base64.b64decode(text, validate=True)
+        return True
+    except Exception:
+        return False
+
+
+def normalize_image_input(image_input: str, upload_timeout: int = 30) -> str:
+    text = image_input.strip()
+    if not text:
+        raise ValueError("图片输入不能为空")
+
+    if text.startswith(("http://", "https://")):
+        return text
+
+    if text.startswith("data:image/") and ";base64," in text:
+        return text
+
+    if os.path.exists(text):
+        return sophnet_tools.upload_oss(text, upload_timeout)
+
+    if _is_likely_base64(text):
+        return text
+
+    raise ValueError("无法识别图片输入：请传入 URL、base64、data URL 或本地文件路径")
+
+
+def _post_chat_completions(payload: dict) -> dict:
+    api_key = sophnet_tools.get_api_key()
+    if not api_key:
+        raise RuntimeError("未配置 API Key，无法进行证件照合规预审")
+    body = json.dumps(payload).encode("utf-8")
+    request = urllib.request.Request(
+        url=API_URL,
+        data=body,
+        method="POST",
+        headers={
+            "Content-Type": "application/json",
+            "Accept": "application/json",
+            "Authorization": f"Bearer {api_key}",
+        },
+    )
+    try:
+        with urllib.request.urlopen(request, timeout=180) as response:
+            raw = response.read().decode("utf-8")
+            return json.loads(raw)
+    except urllib.error.HTTPError as exc:
+        detail = exc.read().decode("utf-8", errors="ignore")
+        raise RuntimeError(f"HTTPError: {exc.code} {exc.reason} | {detail}") from exc
+    except urllib.error.URLError as exc:
+        raise RuntimeError(f"URLError: {exc.reason}") from exc
+
+
+def _extract_message_text(data: dict) -> str:
+    try:
+        return data["choices"][0]["message"]["content"]
+    except (KeyError, IndexError, TypeError):
+        return json.dumps(data, ensure_ascii=False, indent=2)
+
+
+def _strip_json_fence(text: str) -> str:
+    t = text.strip()
+    m = re.match(r"^```(?:json)?\s*([\s\S]*?)\s*```\s*$", t, re.IGNORECASE)
+    if m:
+        return m.group(1).strip()
+    return t
+
+
+def check_id_photo_compliance(image_input: str, *, image_scope: str = "full") -> dict:
+    """
+    返回 dict：compliant, summary, violations, standard_brief, confidence, model；含 raw 供调试。
+    image_scope: \"full\" 整张原图预审；\"cropped\" 裁切后头肩区预审（规范与 system 提示更短，不再强调合照）。
+    """
+    image_url = normalize_image_input(image_input)
+    is_cropped = image_scope == "cropped"
+    standard = ID_PHOTO_STANDARD_BRIEF_CROPPED if is_cropped else ID_PHOTO_STANDARD_BRIEF
+
+    if is_cropped:
+        system = (
+            "你是证件照预处理预审助手。图为**已裁切的单人头肩候选区**，勿编造细节。"
+            "**主要拦截**：佩戴口罩、佩戴墨镜、人脸被遮挡、双耳未露出；**不要**以合照、场景、背景为由拒审。"
+            "violations 措辞优先：「请勿佩戴口罩」「请勿佩戴墨镜」「人脸被遮挡」「双耳未露出」。"
+            "口罩与墨镜须分两条 violation。无硬伤时 compliant=true、violations=[]。"
+            "输出且仅输出 JSON：compliant（布尔）、summary（一至两句中文）、violations（字符串数组）、"
+            "confidence（high/medium/low）。"
+        )
+    else:
+        system = (
+            "你是证件照预处理预审助手。根据用户给出的「规范要点」判断 compliant，不编造图中没有的细节。"
+            "**主要拦截**：合照或无法认定单人主体、佩戴口罩、佩戴墨镜、人脸被遮挡、双耳未露出。"
+            "除此之外宜判定通过；**不要**针对具体场景（如咖啡馆、街道、室内陈设等）或手持物品类别做特化拒审。"
+            "violations 为若干条**简短中文**，命中硬伤时**逐条列出**；"
+            "措辞优先且尽量分列：「疑似合照，无法认定单一人物主体」「请勿佩戴口罩」「请勿佩戴墨镜」「人脸被遮挡」「双耳未露出」。"
+            "若同时存在口罩与墨镜，须分成两条 violation，勿合并为一条。"
+            "无硬伤时 compliant=true 且 violations=[]。"
+            "输出且仅输出一个 JSON，不要 Markdown："
+            '键 compliant（布尔）、summary（一至两句中文）、violations（字符串数组）、'
+            "confidence（high/medium/low）。"
+        )
+    user = (
+        "【规范要点】\n"
+        + standard
+        + "\n\n请审视所附图片，只输出 JSON。"
+    )
+    payload = {
+        "messages": [
+            {"role": "system", "content": system},
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": user},
+                    {"type": "image_url", "image_url": {"url": image_url}},
+                ],
+            },
+        ],
+        "model": MODEL_ID_PHOTO_COMPLIANCE,
+        "stream": False,
+    }
+    data = _post_chat_completions(payload)
+    raw = _extract_message_text(data)
+    try:
+        obj = json.loads(_strip_json_fence(raw))
+    except json.JSONDecodeError as e:
+        raise RuntimeError(f"合规判定返回非 JSON：{raw[:500]}...") from e
+
+    compliant = bool(obj.get("compliant"))
+    violations = obj.get("violations") or []
+    if not isinstance(violations, list):
+        violations = [str(violations)]
+    summary = str(obj.get("summary") or ("符合规范" if compliant else "不符合规范"))
+    return {
+        "compliant": compliant,
+        "summary": summary,
+        "violations": [str(v) for v in violations],
+        "standard_brief": standard,
+        "confidence": str(obj.get("confidence") or "unknown"),
+        "model": MODEL_ID_PHOTO_COMPLIANCE,
+        "raw": raw,
+    }
+
+
+def build_parser():
+    p = argparse.ArgumentParser(description="证件照输入合规预审（视觉模型，本 skill 自包含）")
+    p.add_argument(
+        "--scope",
+        choices=("full", "cropped"),
+        default="full",
+        help="full=整张原图「规范要点」；cropped=头肩裁切后精简要点（不再强调合照）",
+    )
+    p.add_argument(
+        "image",
+        help="图片：http(s) URL、data URL、base64 或本地路径（本地会先上传）",
+    )
+    return p
+
+
+def main():
+    args = build_parser().parse_args()
+    try:
+        report = check_id_photo_compliance(args.image, image_scope=args.scope)
+    except Exception as e:
+        print(str(e), file=sys.stderr)
+        sys.exit(1)
+    out = {k: v for k, v in report.items() if k != "raw"}
+    print(json.dumps(out, ensure_ascii=False))
+    sys.exit(0 if report["compliant"] else 2)
+
+
+if __name__ == "__main__":
+    main()