Refactor feature extraction to handle SURF fallback and add features from regions

This commit updates the feature extraction process to provide a fallback to SIFT when SURF is not available, enhancing robustness. It introduces a new method to extract features from specific regions of a frame, allowing for the addition of features to existing ones. Feedback messages have been improved to reflect these changes, ensuring better user experience and clarity during feature extraction.

croppa/main.py

@@ -49,7 +49,10 @@ class FeatureTracker:
             if self.detector_type == 'SIFT':
                 self.detector = cv2.SIFT_create(nfeatures=self.max_features)
             elif self.detector_type == 'SURF':
-                self.detector = cv2.xfeatures2d.SURF_create(hessianThreshold=400)
+                # SURF requires opencv-contrib-python, fallback to SIFT
+                print("Warning: SURF requires opencv-contrib-python package. Using SIFT instead.")
+                self.detector = cv2.SIFT_create(nfeatures=self.max_features)
+                self.detector_type = 'SIFT'
             elif self.detector_type == 'ORB':
                 self.detector = cv2.ORB_create(nfeatures=self.max_features)
             else:
@@ -107,6 +110,53 @@ class FeatureTracker:
             print(f"Error extracting features from frame {frame_number}: {e}")
             return False
     
+    def extract_features_from_region(self, frame: np.ndarray, frame_number: int, coord_mapper=None) -> bool:
+        """Extract features from a frame and ADD them to existing features"""
+        try:
+            # Convert to grayscale if needed
+            if len(frame.shape) == 3:
+                gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+            else:
+                gray = frame
+            
+            # Extract keypoints and descriptors
+            keypoints, descriptors = self.detector.detectAndCompute(gray, None)
+            
+            if keypoints is None or descriptors is None:
+                return False
+            
+            # Map coordinates back to original frame space if mapper provided
+            if coord_mapper:
+                mapped_positions = []
+                for kp in keypoints:
+                    orig_x, orig_y = coord_mapper(kp.pt[0], kp.pt[1])
+                    mapped_positions.append((int(orig_x), int(orig_y)))
+            else:
+                mapped_positions = [(int(kp.pt[0]), int(kp.pt[1])) for kp in keypoints]
+            
+            # Add to existing features or create new entry
+            if frame_number in self.features:
+                # Append to existing features
+                existing_features = self.features[frame_number]
+                existing_features['keypoints'] = np.concatenate([existing_features['keypoints'], keypoints])
+                existing_features['descriptors'] = np.concatenate([existing_features['descriptors'], descriptors])
+                existing_features['positions'].extend(mapped_positions)
+                print(f"Added {len(keypoints)} features to frame {frame_number} (total: {len(existing_features['positions'])})")
+            else:
+                # Create new features entry
+                self.features[frame_number] = {
+                    'keypoints': keypoints,
+                    'descriptors': descriptors,
+                    'positions': mapped_positions
+                }
+                print(f"Extracted {len(keypoints)} features from frame {frame_number}")
+            
+            return True
+            
+        except Exception as e:
+            print(f"Error extracting features from frame {frame_number}: {e}")
+            return False
+    
     
     
     def get_tracking_position(self, frame_number: int) -> Optional[Tuple[float, float]]:
@@ -1248,8 +1298,11 @@ class VideoEditor:
             self.feature_tracker.auto_tracking and
             self.current_display_frame is not None):
             
+            print(f"DEBUG: Auto-tracking conditions met for frame {self.current_frame}")
+            print(f"DEBUG: tracking_enabled={self.feature_tracker.tracking_enabled}, auto_tracking={self.feature_tracker.auto_tracking}")
             # Only extract if we don't already have features for this frame
             if self.current_frame not in self.feature_tracker.features:
+                print(f"DEBUG: Extracting features for frame {self.current_frame}")
                 # Extract features from the transformed frame (what user sees)
                 # This handles all transformations (crop, zoom, rotation) correctly
                 display_frame = self.apply_crop_zoom_and_rotation(self.current_display_frame)
@@ -1271,6 +1324,8 @@ class VideoEditor:
                         return self._map_screen_to_rotated(screen_x, screen_y)
                     
                     self.feature_tracker.extract_features(display_frame, self.current_frame, coord_mapper)
+            else:
+                print(f"DEBUG: Frame {self.current_frame} already has features, skipping")
 
     def jump_to_previous_marker(self):
         """Jump to the previous tracking marker (frame with tracking points)."""
@@ -1671,10 +1726,11 @@ class VideoEditor:
                         rot_y = orig_y + py
                     return (int(rot_x), int(rot_y))
                 
-                success = self.feature_tracker.extract_features(region_frame, self.current_frame, coord_mapper)
+                # Extract features and add them to existing features
+                success = self.feature_tracker.extract_features_from_region(region_frame, self.current_frame, coord_mapper)
                 if success:
                     count = self.feature_tracker.get_feature_count(self.current_frame)
-                    self.show_feedback_message(f"Extracted {count} features from selected region")
+                    self.show_feedback_message(f"Added features from selected region (total: {count})")
                 else:
                     self.show_feedback_message("Failed to extract features from region")
             else:
@@ -3420,7 +3476,7 @@ class VideoEditor:
             print("  Shift+T: Extract features from current frame")
             print("  g: Toggle auto feature extraction")
             print("  G: Clear all feature data")
-            print("  H: Switch detector (SIFT/SURF/ORB)")
+            print("  H: Switch detector (SIFT/ORB)")
             print("  Shift+Right-click+drag: Extract features from selected region")
             print("  Ctrl+Right-click+drag: Delete features from selected region")
             if len(self.video_files) > 1:
@@ -3727,6 +3783,7 @@ class VideoEditor:
             elif key == ord("g"):
                 # Toggle auto tracking
                 self.feature_tracker.auto_tracking = not self.feature_tracker.auto_tracking
+                print(f"DEBUG: Auto tracking toggled to {self.feature_tracker.auto_tracking}")
                 self.show_feedback_message(f"Auto tracking {'ON' if self.feature_tracker.auto_tracking else 'OFF'}")
                 self.save_state()
             elif key == ord("G"):
@@ -3735,11 +3792,9 @@ class VideoEditor:
                 self.show_feedback_message("Feature tracking data cleared")
                 self.save_state()
             elif key == ord("H"):
-                # Switch detector type (SIFT -> SURF -> ORB -> SIFT)
+                # Switch detector type (SIFT -> ORB -> SIFT) - SURF not available
                 current_type = self.feature_tracker.detector_type
                 if current_type == 'SIFT':
-                    new_type = 'SURF'
-                elif current_type == 'SURF':
                     new_type = 'ORB'
                 elif current_type == 'ORB':
                     new_type = 'SIFT'