Update app.py

app.py

@@ -64,24 +64,42 @@ def detect_face_local(image_pil):
     if not face_image_processor or not face_detection_model or FACE_LABEL_ID == -1:
         return None, "Face detection model not loaded or configured properly."
 
+    print(f"Detecting face with FACE_LABEL_ID: {FACE_LABEL_ID}")
+    detection_threshold = 0.4 # <<-- TRY LOWERING THIS (e.g., 0.5, 0.4, 0.3)
+    print(f"Using detection threshold: {detection_threshold}")
+
     try:
         inputs = face_image_processor(images=image_pil, return_tensors="pt")
-        with torch.no_grad(): # Important for inference
+        with torch.no_grad():
             outputs = face_detection_model(**inputs)
 
         target_sizes = torch.tensor([image_pil.size[::-1]])
-        results = face_image_processor.post_process_object_detection(outputs, threshold=0.7, target_sizes=target_sizes)[0] # Threshold
+        # Setting a lower threshold for post-processing here
+        results = face_image_processor.post_process_object_detection(
+            outputs, threshold=detection_threshold, target_sizes=target_sizes
+        )[0]
 
         best_box = None
-        max_score = 0
+        max_score = 0 # We will still pick the best one above the (now lower) threshold
+
+        print(f"Detection results: {len(results['scores'])} detections before filtering by label.")
+        detected_items_for_label = []
 
         for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
-            if label.item() == FACE_LABEL_ID: # Use .item() to get Python int from tensor
-                if score.item() > max_score:
-                    max_score = score.item()
+            current_score = score.item()
+            current_label = label.item()
+            print(f"  - Detected item: Label {current_label}, Score {current_score:.2f}")
+            if current_label == FACE_LABEL_ID:
+                detected_items_for_label.append({'score': current_score, 'box': box.tolist()})
+                if current_score > max_score:
+                    max_score = current_score
                     best_box = box.tolist()
+        
+        print(f"Found {len(detected_items_for_label)} items matching FACE_LABEL_ID {FACE_LABEL_ID} with scores: {[item['score'] for item in detected_items_for_label]}")
 
         if best_box:
+            print(f"Selected best box with score: {max_score:.2f}")
+            # Add a small padding to the bounding box
             padding_w = (best_box[2] - best_box[0]) * 0.15 # 15% padding width
             padding_h = (best_box[3] - best_box[1]) * 0.15 # 15% padding height
 
@@ -89,13 +107,26 @@ def detect_face_local(image_pil):
             ymin = max(0, best_box[1] - padding_h)
             xmax = min(image_pil.width, best_box[2] + padding_w)
             ymax = min(image_pil.height, best_box[3] + padding_h)
+            
+            # Ensure cropped dimensions are valid
+            if xmax <= xmin or ymax <= ymin:
+                print(f"Warning: Invalid crop dimensions after padding. Original box: {best_box}. Padded: ({xmin},{ymin},{xmax},{ymax})")
+                # Fallback to original box if padding made it invalid
+                xmin, ymin, xmax, ymax = best_box[0], best_box[1], best_box[2], best_box[3]
+                if xmax <= xmin or ymax <= ymin: # If original box itself is invalid
+                     return None, "Detected box has invalid dimensions."
 
             cropped_image = image_pil.crop((xmin, ymin, xmax, ymax))
             return cropped_image, None
         else:
-            return None, "No face/person detected with sufficient confidence."
+            if len(detected_items_for_label) > 0:
+                 return None, f"Faces detected but scores too low (max score: {max_score:.2f} with threshold {detection_threshold}). Try a clearer image or different pose."
+            else:
+                 return None, f"No face/person detected with sufficient confidence (threshold {detection_threshold}). Ensure face is clear and well-lit."
     except Exception as e:
         print(f"Error during local face detection: {e}")
+        import traceback
+        traceback.print_exc() # Print full traceback for debugging
         return None, f"Error during face detection: {str(e)}"