fix : face not detecting

src/models.py

@@ -109,23 +109,30 @@ class AIModelManager:
 
         # ── Face Lane: InsightFace (YuNet + ArcFace) ─────────────
         self.face_app = None
+        print(f"🔍 INSIGHTFACE_AVAILABLE = {INSIGHTFACE_AVAILABLE}")
         if INSIGHTFACE_AVAILABLE:
             try:
-                # buffalo_sc = small+fast model (CPU optimised)
-                # buffalo_l  = large+accurate (use if GPU available)
+                import insightface
+                print(f"🔍 InsightFace version: {insightface.__version__}")
                 model_name = "buffalo_l" if self.device == "cuda" else "buffalo_sc"
+                print(f"🔍 Loading InsightFace model: {model_name}")
                 self.face_app = FaceAnalysis(name=model_name)
-                # det_size controls detection resolution — larger = finds smaller faces
                 self.face_app.prepare(
                     ctx_id=0 if self.device == "cuda" else -1,
                     det_size=(640, 640),
                 )
-                print(f"✅ InsightFace ({model_name}) loaded — ArcFace face lane active")
+                # Test with a blank image to confirm models loaded
+                import numpy as _np
+                test_img = _np.zeros((112, 112, 3), dtype=_np.uint8)
+                _ = self.face_app.get(test_img)
+                print(f"✅ InsightFace ({model_name}) loaded — ArcFace face lane ACTIVE")
             except Exception as e:
-                print(f"⚠️  InsightFace init failed: {e} — face lane disabled")
+                import traceback
+                print(f"❌ InsightFace init FAILED: {e}")
+                print(traceback.format_exc())
                 self.face_app = None
         else:
-            print("⚠️  InsightFace not available — install: pip install insightface onnxruntime")
+            print("❌ InsightFace NOT installed — run: pip install insightface onnxruntime")
 
         self._cache          = {}
         self._cache_maxsize  = 128
@@ -176,9 +183,11 @@ class AIModelManager:
         }
         """
         if self.face_app is None:
+            print("⚠️  face_app is None — InsightFace not loaded!")
             return []
 
         try:
+            print(f"🔍 Running InsightFace on image shape: {img_np.shape}")
             # InsightFace expects BGR numpy array
             if img_np.shape[2] == 3 and img_np.dtype == np.uint8:
                 bgr = img_np[..., ::-1].copy()  # RGB → BGR
@@ -187,6 +196,7 @@ class AIModelManager:
 
             with self._face_lock:
                 faces = self.face_app.get(bgr)
+            print(f"🔍 InsightFace raw detection: {len(faces)} faces found")
             results = []
 
             for idx, face in enumerate(faces):
@@ -281,23 +291,16 @@ class AIModelManager:
                     extracted.append(fr)
 
         # ── OBJECT LANE ──────────────────────────────────────────
-        # ALWAYS runs — stores full image + YOLO crops in enterprise-objects
-        # This is critical: even when faces found, the full image must be
-        # stored in enterprise-objects so face search can retrieve it!
-        #
-        # Strategy (matches cloud_db.py):
-        #   - full image ALWAYS included as first crop
-        #   - YOLO person crops SKIPPED when faces found (avoid duplication)
-        #   - other object crops always included
-        crops_pil    = [_resize_pil(original_pil, MAX_IMAGE_SIZE)]  # ALWAYS include full image
+        # Always run object lane — even if faces found
+        # (image may contain both people and objects)
+        crops_pil    = [_resize_pil(original_pil, MAX_IMAGE_SIZE)]  # full-image always
         yolo_results = self.yolo(image_path, conf=0.5, verbose=False)
 
         for r in yolo_results:
             if r.masks is not None:
                 for seg_idx, mask_xy in enumerate(r.masks.xy):
                     cls_id = int(r.boxes.cls[seg_idx].item())
-                    # Skip YOLO person crops when face lane handled faces
-                    # (avoids storing low-quality person crops redundantly)
+                    # Skip person crops if face lane already handled them
                     if faces_found and cls_id == YOLO_PERSON_CLASS_ID:
                         print("🔵 PERSON crop skipped — face lane active")
                         continue
@@ -324,7 +327,6 @@ class AIModelManager:
             if len(crops_pil) >= MAX_CROPS + 1:
                 break
 
-        # Embed all crops — full image is ALWAYS first in the list
         crops = [_resize_pil(c, MAX_IMAGE_SIZE) for c in crops_pil]
         print(f"🧠 Embedding {len(crops)} object crop(s) in one batch …")
         obj_vecs = self._embed_crops_batch(crops)