apps/worker/worker_app/tasks/vision/assembler.py

@@ -420,7 +420,7 @@ def _assemble_v4(idx: int, fj: dict, ocr_texts: list[str]) -> dict[str, Any]:
 
     # 人物类
     if vtype == "person" or has_person:
-        # 取第一个人物信息（v4 schema人物信息在顶层）
+        # 取第一个人物信息（v5 schema人物信息在顶层）
         person_info = fj
         # 兼容people嵌套
         ppl = fj.get("people")
@@ -429,58 +429,75 @@ def _assemble_v4(idx: int, fj: dict, ocr_texts: list[str]) -> dict[str, Any]:
             has_person = True
 
         portrait_prompt = _build_portrait_prompt_from_v4(person_info)
-        name = person_info.get("upper_wear") or "人物穿搭"
-        if "连衣裙" in name:
-            pass
+        outfit_style = person_info.get("outfit_style") or ""
+        upper = person_info.get("upper_wear") or ""
+        lower = person_info.get("lower_wear") or ""
+        dress = person_info.get("dress_wear") or ""
+        outer = person_info.get("outerwear") or ""
+        if dress:
+            name = str(dress)[:25]
+        elif outer and upper:
+            name = f"{outer}+{upper}"[:30]
+        elif upper:
+            name = (str(upper) + (f"+{lower}" if lower else ""))[:30]
         else:
-            lower = person_info.get("lower_wear") or ""
-            if lower:
-                name = f"{name}+{lower}"
+            name = "人物穿搭"
         brand = "无法判断"
-        category = "服饰"
-        outfit_parts = []
-        for k in ("upper_wear", "lower_wear", "dress_color", "upper_color", "lower_color", "outfit_style"):
+        category = "人物穿搭"
+        # appearance: 外套+上衣+下装/裙+鞋+包+发型+妆容
+        app_parts = []
+        for k in ("outerwear", "upper_wear", "lower_wear", "dress_wear", "shoes", "bag", "hairstyle", "makeup"):
             v = person_info.get(k)
-            if v and v not in ("null", None):
-                outfit_parts.append(str(v))
-        appearance = "、".join(outfit_parts) if outfit_parts else "人像穿搭整体造型"
-        # key_features: 穿搭特征+配饰
+            if v and v not in ("null", None, "无明显妆容"):
+                app_parts.append(str(v))
+        appearance = "；".join(app_parts) if app_parts else "人像穿搭整体造型"
+        # key_features: 服装+配饰+拍摄信息
         kf = []
         for k in (
+            "outfit_style",
             "upper_wear",
             "lower_wear",
-            "upper_color",
-            "lower_color",
+            "dress_wear",
+            "outerwear",
+            "shoes",
+            "bag",
             "hairstyle",
             "expression",
             "pose",
-            "outfit_style",
         ):
             v = person_info.get(k)
             if v and v not in ("null", None, "无法判断"):
                 kf.append(str(v))
         acc = person_info.get("accessories") or []
         if isinstance(acc, list):
-            kf.extend(str(a) for a in acc if a)
+            for a in acc:
+                if a and str(a) not in kf:
+                    kf.append(str(a))
+        elif isinstance(acc, str) and acc:
+            kf.append(acc)
+        for k in ("shot_type", "camera_angle", "lighting", "atmosphere"):
+            v = person_info.get(k)
+            if v and v not in ("null", None):
+                kf.append(str(v))
         if text_on_package:
-            kf.append(f"画面文字: {'/'.join(text_on_package[:3])}")
-        kf = kf[:6] or ["无法判断"]
-        summary = (person_info.get("outfit_style") or "") + (person_info.get("upper_wear") or "穿搭")
-        if not summary or summary == "穿搭":
+            kf.append(f"文字:{'/'.join(text_on_package[:3])}")
+        kf = kf[:8] or ["无法判断"]
+        summary = (outfit_style + " " if outfit_style and outfit_style not in name else "") + name[:25]
+        if not summary.strip():
             summary = "人物穿搭"
         return {
             "name": name[:30],
             "brand": brand,
             "category": category,
-            "appearance": appearance,
+            "appearance": appearance[:400],
             "packaging": "人物形象无包装",
             "text_on_package": text_on_package,
             "key_features": kf,
             "scene": scene,
             "mood": mood,
-            "portrait_prompt": portrait_prompt,
-            "summary": summary[:40],
-            "_source": "v2_fast_json_v4",
+            "portrait_prompt": portrait_prompt[:300],
+            "summary": summary[:50],
+            "_source": "v2_fast_json_v5",
         }
 
     # 商品类

tests/unit/test_vision_v4_prompt.py

@@ -137,18 +137,18 @@ V4_PERSON: dict[str, Any] = {
 def test_assemble_v4_person() -> None:
     r = assembler.assemble_result(0, V4_PERSON, [])
     assert REQUIRED_KEYS <= set(r.keys())
-    assert r["category"] == "服饰"
+    assert r["category"] == "人物穿搭"
+    assert r["_source"] == "v2_fast_json_v5"
     assert "T恤" in r["name"]
-    assert "阔腿裤" in r["name"]
     assert "年轻女性" in r["portrait_prompt"]
     assert "项链" in r["portrait_prompt"]
-    assert isinstance(r["key_features"], list) and len(r["key_features"]) <= 6
+    assert isinstance(r["key_features"], list) and len(r["key_features"]) <= 8
 
 
 def test_assemble_v4_person_people_nested() -> None:
     fj = {"type": "person", "people": {**V4_PERSON, "has_person": True}}
     r = assembler.assemble_result(0, fj, [])
-    assert r["category"] == "服饰"
+    assert r["category"] == "人物穿搭"
     assert "年轻女性" in r["portrait_prompt"]