- Add debug mode support for encoding statistics in API responses
- Debug info includes encoding length, min/max/mean/std, and first 10 values
- Frontend logs encoding stats to browser console when debug enabled
- Identify page enables debug mode by default
- Implement distance-based confidence thresholds for stricter matching
- Borderline distances require higher confidence (70-95% vs 50%)
- Applied when use_distance_based_thresholds=True (auto-match)
- Reduces false positives for borderline matches
- Dual tolerance system for auto-match
- Default tolerance 0.6 for regular browsing (more lenient)
- Run auto-match button uses 0.5 tolerance with distance-based thresholds (stricter)
- Auto-accept threshold updated to 85% (from 70%)
- Enhance pose detection with single-eye detection
- Profile threshold reduced from 30° to 15° (stricter)
- Detect single-eye visibility for extreme profile views
- Infer profile direction from landmark visibility
- Improved face width threshold (20px vs 10px)
- Clean up debug code
- Remove test photo UUID checks from production code
- Remove debug print statements
- Replace print statements with proper logging
backend/config.py - Added MIN_AUTO_MATCH_FACE_SIZE_RATIO = 0.005
backend/services/face_service.py - Multiple changes:
Added load_face_encoding() function (supports float32 and float64)
Added _calculate_face_size_ratio() function
Updated find_similar_faces() to filter small faces
Updated find_auto_match_matches() to exclude small reference faces
Fixed reference face quality calculation (use actual quality, not hardcoded 0.5)
Fixed duplicate detection (exclude faces from same photo)
Updated confidence threshold from 40% to 50%
Updated confidence calibration (moderate version)
backend/api/faces.py - Updated default tolerance to 0.5 for auto-match endpoints
backend/schemas/faces.py - Updated default tolerance to 0.5
admin-frontend/src/pages/AutoMatch.tsx - Updated default tolerance to 0.5
admin-frontend/src/api/faces.ts - Added tolerance parameter support
- Added new logging scripts for quick access to service logs and troubleshooting.
- Updated job streaming API to support authentication via query parameters for EventSource.
- Improved photo upload process to capture and validate EXIF dates and original modification times.
- Enhanced error handling for file uploads and EXIF extraction failures.
- Introduced new configuration options in ecosystem.config.js to prevent infinite crash loops.
- Introduced the exifread library to improve reliability in extracting EXIF date from images.
- Updated the extract_exif_date function to prioritize exifread for date extraction.
- Enhanced logging for successful and failed date extractions.
- Added validation for extracted dates to ensure they fall within a valid range.
- Improved logging for EXIF date extraction process, including success and failure messages.
- Updated methods to access EXIF data using modern and deprecated APIs.
- Added validation for extracted dates to ensure they fall within a valid range.
- Enhanced error handling for various exceptions during EXIF data access and date parsing.
- Added media_type to PhotoSearchResult interface to distinguish between images and videos.
- Updated PhotoViewer component to support video playback, including URL handling and preloading logic.
- Modified openPhoto function in Search page to open videos correctly.
- Enhanced backend API to serve video files with range request support for better streaming experience.
These changes improve the user experience by allowing seamless viewing of both images and videos in the application.
- Add database connection health checks every 10 photos
- Add session refresh logic to recover from connection errors
- Improve error handling for database disconnections/timeouts
- Add explicit image cleanup to prevent memory leaks
- Add connection error detection throughout processing pipeline
- Gracefully handle database connection failures instead of crashing
Fixes issue where server would crash during long-running photo processing
tasks when database connections were lost or timed out.
This commit introduces a new `pytest.ini` configuration file for backend tests, specifying test discovery patterns and output options. Additionally, the CI workflow is updated to set an environment variable that prevents DeepFace and TensorFlow from loading during tests, avoiding illegal instruction errors on certain CPUs. The face service and pose detection modules are modified to conditionally import DeepFace and RetinaFace based on this environment variable, enhancing test reliability. These changes improve the testing setup and contribute to a more robust CI process.
This commit introduces a new `PrismaCompatibleDate` type to ensure compatibility with Prisma's SQLite driver by storing dates in a DateTime format. Additionally, the `extract_exif_date`, `extract_video_date`, and `extract_photo_date` functions are updated to include validation checks that reject future dates and dates prior to 1900, enhancing data integrity during photo and video metadata extraction.
This commit enhances the face identification process by adding validation checks for person ID and names, ensuring that users provide necessary information before proceeding. It also introduces detailed logging for better debugging and user feedback during the identification process. Additionally, error handling is improved to provide user-friendly messages in case of failures, enhancing the overall user experience.
This commit introduces a new `CANCELLED` status to the job management system, allowing users to cancel ongoing jobs. The frontend is updated to handle job cancellation requests, providing user feedback during the cancellation process. Additionally, the backend is enhanced to manage job statuses more effectively, ensuring that jobs can be marked as cancelled and that appropriate messages are displayed to users. This improvement enhances the overall user experience by providing better control over job processing.
This commit introduces a new function, `validate_date_taken`, to ensure that the date taken for photos is valid before being saved. The function checks for valid date objects, prevents future dates, and filters out dates that are too old. The photo import process is updated to utilize this validation, enhancing data integrity and preventing corrupted date data from being stored. Additionally, the `date_added` field is explicitly set to the current UTC time to ensure validity.
This commit introduces a new function in the `install.sh` script to download DeepFace model weights, enhancing the setup process for users. The function checks for the presence of DeepFace and attempts to download the ArcFace model weights, providing fallback options and user-friendly messages for manual download if automatic attempts fail. This improvement streamlines the initial configuration for facial recognition capabilities in the application.
This commit introduces several new analysis documents, including Auto-Match Load Performance Analysis, Folder Picker Analysis, Monorepo Migration Summary, and various performance analysis documents. Additionally, the installation scripts are updated to reflect changes in backend service paths, ensuring proper integration with the new backend structure. These enhancements provide better documentation and streamline the setup process for users.
