Update app.py

app.py

@@ -402,8 +402,15 @@ class ForgeryDetector:
         # Extract regions
         regions = self.region_extractor.extract(refined_mask, prob_map_resized, original_image)
         
+        print(f"[DEBUG] Regions extracted: {len(regions)}")
+        if len(regions) > 0:
+            print(f"[DEBUG] Region areas: {[r['area'] for r in regions]}")
+            print(f"[DEBUG] Region confidences: {[r.get('confidence', 0) for r in regions]}")
+        
         # Classify regions
         results = []
+        classified_count = 0
+        rejected_count = 0
         for region in regions:
             # Get decoder features and handle shape
             df = decoder_features[0].cpu()  # Get first decoder feature
@@ -446,19 +453,36 @@ class ForgeryDetector:
             elif current_features > expected_features:
                 features = features[:, :expected_features]
             
-            # Classify
-            predictions, confidences = self.classifier.predict(features)
-            forgery_type = int(predictions[0])
-            confidence = float(confidences[0])
+            # Classify - get probabilities for all classes
+            # Temporarily access model directly to get full probabilities
+            features_scaled = self.classifier.scaler.transform(features)
+            probabilities = self.classifier.model.predict(features_scaled)[0]  # Shape: (3,)
+            
+            forgery_type = int(probabilities.argmax())
+            confidence = float(probabilities.max())
+            
+            # Log all class probabilities for debugging
+            prob_str = ", ".join([f"{CLASS_NAMES[i]}: {probabilities[i]:.3f}" for i in range(3)])
+            print(f"[DEBUG] Region {region['region_id']}: {CLASS_NAMES[forgery_type]} (confidence: {confidence:.3f})")
+            print(f"        All probabilities: {prob_str}")
             
             # Lower confidence threshold to detect more regions
             if confidence > 0.5:
+                classified_count += 1
                 results.append({
                     'region_id': region['region_id'],
                     'bounding_box': region['bounding_box'],
                     'forgery_type': CLASS_NAMES[forgery_type],
                     'confidence': confidence
                 })
+            else:
+                rejected_count += 1
+                print(f"  -> REJECTED (confidence {confidence:.3f} < 0.5)")
+        
+        print(f"[DEBUG] Classification summary:")
+        print(f"  - Total regions: {len(regions)}")
+        print(f"  - Classified: {classified_count}")
+        print(f"  - Rejected: {rejected_count}")
         
         # Create visualization
         overlay = self._create_overlay(original_image, results)