dave/py-media-grader
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refactor feature extraction to use transformed frames in VideoEditor

Browse Source 
This commit updates the feature extraction process to utilize the transformed frames that users see, rather than the original frames. It includes a new method for mapping coordinates from transformed frames back to original coordinates, ensuring accurate tracking. Additionally, feedback messages have been improved to reflect the success or failure of feature extraction based on the visible area.
This commit is contained in:
PhatPhuckDave
2025-09-26 13:17:13 +02:00
parent 171155e528
commit c50234f5c1
1 changed files with 27 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
croppa/main.py
View File

@@ -48,13 +48,10 @@ class FeatureTracker:
try:
if self.detector_type == 'SIFT':
self.detector = cv2.SIFT_create(nfeatures=self.max_features)
self.matcher = cv2.BFMatcher()
elif self.detector_type == 'SURF':
self.detector = cv2.xfeatures2d.SURF_create(hessianThreshold=400)
self.matcher = cv2.BFMatcher()
elif self.detector_type == 'ORB':
self.detector = cv2.ORB_create(nfeatures=self.max_features)
self.matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
else:
raise ValueError(f"Unknown detector type: {self.detector_type}")
except Exception as e:
@@ -62,7 +59,6 @@ class FeatureTracker:
# Fallback to ORB
self.detector_type = 'ORB'
self.detector = cv2.ORB_create(nfeatures=self.max_features)
self.matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
def set_detector_type(self, detector_type: str):
"""Change detector type and reinitialize"""
@@ -1239,9 +1235,10 @@ class VideoEditor:
# Only extract if we don't already have features for this frame
if self.current_frame not in self.feature_tracker.features:
# Extract features from the original frame (before transformations)
# This ensures features are in the original coordinate system
self.feature_tracker.extract_features(self.current_display_frame, self.current_frame)
# Extract features from the transformed frame (what user sees after crop/zoom/rotation)
display_frame = self.apply_crop_zoom_and_rotation(self.current_display_frame)
if display_frame is not None:
self.feature_tracker.extract_features(display_frame, self.current_frame)
def jump_to_previous_marker(self):
"""Jump to the previous tracking marker (frame with tracking points)."""
@@ -1456,14 +1453,23 @@ class VideoEditor:
h = min(h, rot_h - y)
return (x, y, w, h)
def _map_transformed_to_original_coords(self, x, y):
"""Map coordinates from transformed frame back to original frame coordinates."""
# This is a simplified mapping - in practice, we'd need to reverse all transformations
# For now, just return the coordinates as-is since the tracking system expects
# coordinates in the original frame space, but we're extracting from transformed frame
# This is a limitation that needs proper coordinate transformation
return (x, y)
def _get_interpolated_tracking_position(self, frame_number):
"""Linear interpolation in ROTATED frame coords. Returns (rx, ry) or None."""
# First try feature tracking if enabled
if self.feature_tracker.tracking_enabled:
feature_pos = self.feature_tracker.get_tracking_position(frame_number)
if feature_pos:
# Features are extracted from original frames, so coordinates are already correct
return feature_pos
# Features are extracted from transformed frame, need to map back to original
orig_x, orig_y = self._map_transformed_to_original_coords(feature_pos[0], feature_pos[1])
return (orig_x, orig_y)
# Fall back to manual tracking points
if not self.tracking_points:
@@ -3498,15 +3504,18 @@ class VideoEditor:
elif key == ord("T"):
# Extract features from current frame (Shift+T)
if not self.is_image_mode and self.current_display_frame is not None:
# Extract features from the original frame (before transformations)
# This ensures features are in the original coordinate system
success = self.feature_tracker.extract_features(self.current_display_frame, self.current_frame)
if success:
count = self.feature_tracker.get_feature_count(self.current_frame)
self.show_feedback_message(f"Extracted {count} features from frame {self.current_frame}")
# Extract features from the transformed frame (what user sees after crop/zoom/rotation)
display_frame = self.apply_crop_zoom_and_rotation(self.current_display_frame)
if display_frame is not None:
success = self.feature_tracker.extract_features(display_frame, self.current_frame)
if success:
count = self.feature_tracker.get_feature_count(self.current_frame)
self.show_feedback_message(f"Extracted {count} features from transformed frame")
else:
self.show_feedback_message("Failed to extract features")
self.save_state()
else:
self.show_feedback_message("Failed to extract features")
self.save_state()
self.show_feedback_message("No display frame available")
else:
self.show_feedback_message("No frame data available")
elif key == ord("g"):
@@ -3522,6 +3531,7 @@ class VideoEditor:
elif key == ord("H"):
# Switch detector type (SIFT -> SURF -> ORB -> SIFT)
current_type = self.feature_tracker.detector_type
print(f"Current detector type: {current_type}")
if current_type == 'SIFT':
new_type = 'SURF'
elif current_type == 'SURF':