Update src/models.py

src/models.py

@@ -52,67 +52,52 @@ class AIModelManager:
 
         return object_vec.flatten().numpy()
 
-    def process_image(self, image_path: str, is_query=False):
-        """
-        Master function: Extracts EVERY face and EVERY non-human object from an image.
-
-        Key design decisions:
-        - Face lane runs first and tags every face with its bounding box area.
-        - Only faces above MIN_FACE_AREA are indexed (filters background/tiny faces).
-        - For queries, ALL detected faces are used (not just the first one).
-        - Object lane SKIPS any YOLO detection whose class is 'person', so humans
-          never pollute the object index when faces were already found.
-        - If NO faces are found at all, humans caught by YOLO DO go into the object
-          lane (as a fallback for silhouettes, backs-of-head, full body shots etc.)
-        """
+    # Change the function signature to accept detect_faces
+    def process_image(self, image_path: str, is_query=False, detect_faces=True):
         extracted_vectors = []
         original_img_pil = Image.open(image_path).convert('RGB')
         img_np = np.array(original_img_pil)
         img_h, img_w = img_np.shape[:2]
 
-        faces_were_found = False  # Track whether Lane 1 found anything usable
+        faces_were_found = False  
 
         # ==========================================
-        # LANE 1: THE FACE LANE
+        # LANE 1: THE FACE LANE (NOW TOGGLEABLE)
         # ==========================================
-        try:
-            face_objs = DeepFace.represent(
-                img_path=img_np,
-                model_name="GhostFaceNet",
-                detector_backend="retinaface",
-                enforce_detection=True,
-                align=True
-            )
-
-            for index, face in enumerate(face_objs):
-                # --- BUG FIX 5: Filter out tiny/background faces ---
-                facial_area = face.get("facial_area", {})
-                fw = facial_area.get("w", img_w)
-                fh = facial_area.get("h", img_h)
-                face_area_px = fw * fh
-
-                if face_area_px < MIN_FACE_AREA:
-                    print(f"🟡 FACE {index+1} SKIPPED: Too small ({fw}x{fh}px = {face_area_px}px²) — likely background noise.")
-                    continue
-
-                face_vec = torch.tensor([face["embedding"]])
-                face_vec = F.normalize(face_vec, p=2, dim=1)
-
-                extracted_vectors.append({
-                    "type": "face",
-                    "vector": face_vec.flatten().numpy()
-                })
-                faces_were_found = True
-                print(f"🟢 FACE {index+1} EXTRACTED: {fw}x{fh}px — Added to Face Index.")
-
-                # --- BUG FIX 2: For queries, do NOT break — search with ALL faces ---
-                # The calling code in main.py already loops over all returned vectors,
-                # so returning multiple face vectors means we search for every person
-                # in a group photo query simultaneously.
-                # (is_query flag is kept as parameter for future use / logging only)
-
-        except ValueError:
-            print("🟠 NO FACES DETECTED -> Falling back to Object Lane for any humans.")
+        if detect_faces:
+            try:
+                print("Running heavy face detection...")
+                face_objs = DeepFace.represent(
+                    img_path=img_np,
+                    model_name="GhostFaceNet",
+                    detector_backend="retinaface",
+                    enforce_detection=True,
+                    align=True
+                )
+
+                for index, face in enumerate(face_objs):
+                    facial_area = face.get("facial_area", {})
+                    fw = facial_area.get("w", img_w)
+                    fh = facial_area.get("h", img_h)
+                    face_area_px = fw * fh
+
+                    if face_area_px < MIN_FACE_AREA:
+                        continue
+
+                    face_vec = torch.tensor([face["embedding"]])
+                    face_vec = F.normalize(face_vec, p=2, dim=1)
+
+                    extracted_vectors.append({
+                        "type": "face",
+                        "vector": face_vec.flatten().numpy()
+                    })
+                    faces_were_found = True
+
+            except ValueError:
+                print("🟠 NO FACES DETECTED -> Falling back to Object Lane.")
+        else:
+            print("⏩ FAST MODE: Skipping Face Detection Lane entirely.")
+
 
         # ==========================================
         # LANE 2: THE OBJECT LANE