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: Removed det and rec from ignored directories to include new detection and recognition modules.
• Enhanced Documentation:
  • Added inference example to README.md and README_tw.md.
  • Updated links to papers in README.md and README_tw.md to point to English versions.
  • Added a PyPI download badge to README.md and README_tw.md.
• Refactored MRZScanner Class:
  • Added support for different model types (spotting, detection, recognition, and two_stage).
  • Improved modularity by splitting detection and recognition into separate components.
• Dependency Update: Updated capybara_docsaid requirement 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 detection and recognition inference modules.
  • Expanded test coverage for MRZScanner class.
  • Added mock tests for automatic model download and preprocessing logic.

Full Changelog: 0.3.2...1.0.6