feat(vision): 实现 mingxi-vision 推理服务核心功能

- engine/defect_classes: HRIPCB 6类缺陷定义 + 4类扩展 + severity排序 - engine/base: DefectBox / DetectResult 数据类 + BaseInferenceEngine 抽象基类 - engine/ultralytics_adapter: YOLOv8 .pt 推理适配器 + warmup - engine/annotator: 按 severity 着色的标注图生成 + base64 编码 - engine/loader: 单例管理 + asyncio.Lock 推理串行化 + 热重载支持 - config: Pydantic v1 BaseSettings，读取 .env - schema: DetectResponse / HealthResponse / ReloadRequest 等响应模型 - api/detect: POST /api/detect，内存读图，推理锁保护 - api/health: GET /api/health - api/model: POST /api/model/reload 热重载接口 - main: FastAPI lifespan 启动加载模型测试通过：/api/health 200，/api/detect 空图返回 pass，标注图 base64 正常 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-24 21:18:34 +08:00
parent e3c2cb2d89
commit 68e4f444c5
12 changed files with 406 additions and 0 deletions
@@ -0,0 +1,56 @@
+import time
+from pathlib import Path
+
+import numpy as np
+
+from .base import BaseInferenceEngine, DefectBox, DetectResult
+from .defect_classes import get_class_info
+
+
+class UltralyticsAdapter(BaseInferenceEngine):
+
+    def __init__(self, model_path: str):
+        from ultralytics import YOLO
+        self._model_path = str(model_path)
+        self._model = YOLO(self._model_path)
+        self._version = Path(model_path).stem
+
+    def detect(self, image: np.ndarray, conf: float = 0.45) -> DetectResult:
+        h, w = image.shape[:2]
+        t0 = time.perf_counter()
+        results = self._model(image, conf=conf, verbose=False)
+        duration_ms = (time.perf_counter() - t0) * 1000
+
+        defects = []
+        for r in results:
+            if r.boxes is None:
+                continue
+            for box in r.boxes:
+                class_id = int(box.cls[0])
+                confidence = float(box.conf[0])
+                xyxy = box.xyxy[0].tolist()
+                info = get_class_info(class_id)
+                defects.append(DefectBox(
+                    class_id=class_id,
+                    class_name=info["name"],
+                    class_name_zh=info["zh"],
+                    confidence=round(confidence, 4),
+                    severity=info["severity"],
+                    box_xyxy=[round(v, 1) for v in xyxy],
+                ))
+
+        return DetectResult(
+            defects=defects,
+            duration_ms=round(duration_ms, 1),
+            image_width=w,
+            image_height=h,
+            model_version=self._version,
+        )
+
+    def warmup(self) -> None:
+        dummy = np.zeros((640, 640, 3), dtype=np.uint8)
+        self.detect(dummy, conf=0.45)
+
+    @property
+    def model_version(self) -> str:
+        return self._version