genki
|
f51cd4c9ba
|
feat(training): Add parallel face detection, exclude functionality, and optimize image replacement
PERFORMANCE IMPROVEMENTS:
- Parallel face detection: 30 images now process in ~5s (was ~45s) via batched async processing
- Optimized replace: Only rescans single replaced image instead of entire set
- Memory efficient: Proper bitmap recycling in finally blocks prevents memory leaks
NEW FEATURES:
- Exclude/Include buttons: One-click removal of bad training images with instant UI feedback
- Excluded image styling: Gray overlay, disabled buttons, clear "Excluded" status
- Smart button visibility: Hide Replace/Pick Face when image excluded
- Progress tracking: Real-time callbacks during face detection scan
BUG FIXES:
- Fixed bitmap.recycle() timing to prevent corrupted face crops
- Fixed FaceDetectionHelper to recycle bitmaps only after cropping complete
- Enhanced TrainViewModel with exclude tracking and efficient state updates
UI UPDATES:
- Added ImageStatus.EXCLUDED enum value
- Updated ScanResultsScreen with exclude/include action buttons
- Enhanced color schemes for all 5 image states (Valid, Multiple, None, Error, Excluded)
- Added RemoveCircle icon for excluded images
FILES MODIFIED:
- FaceDetectionHelper.kt: Parallel processing, proper bitmap lifecycle
- TrainViewModel.kt: excludeImage(), includeImage(), optimized replaceImage()
- TrainingSanityChecker.kt: Exclusion support, progress callbacks
- ScanResultsScreen.kt: Complete exclude UI implementation
TESTING:
- 9x faster initial scan (45s → 5s for 30 images)
- 45x faster replace (45s → 1s per image)
- Instant exclude/include (<0.1s UI update)
- Minimum 15 images required for training
|
2026-01-10 09:44:26 -05:00 |
|