Release 1.0.6
Release Note
New Features
- Two-stage MRZ Detection & Recognition: Added support for a two-stage model that includes both detection and recognition modules for improved MRZ extraction accuracy.
Improvements
- Updated
.gitignore: Removeddetandrecfrom ignored directories to include new detection and recognition modules. - Enhanced Documentation:
- Added inference example to
README.mdandREADME_tw.md. - Updated links to papers in
README.mdandREADME_tw.mdto point to English versions. - Added a PyPI download badge to
README.mdandREADME_tw.md.
- Added inference example to
- Refactored
MRZScannerClass:- Added support for different model types (
spotting,detection,recognition, andtwo_stage). - Improved modularity by splitting detection and recognition into separate components.
- Added support for different model types (
- Dependency Update: Updated
capybara_docsaidrequirement to>=0.8.0.
Bug Fixes
- Postprocessing Fixes:
- Improved MRZ text validation and handling of incorrect line lengths.
- Fixed incorrect error handling for TD1 and TD3 postprocessing.
- Backend Fixes:
- Adjusted normalization in image preprocessing to avoid unnecessary precision loss.
- Fixed incorrect padding calculations in image preprocessing for detection models.
Testing
- Unit Tests Added:
- New tests for
detectionandrecognitioninference modules. - Expanded test coverage for
MRZScannerclass. - Added mock tests for automatic model download and preprocessing logic.
- New tests for
Full Changelog: 0.3.2...1.0.6