- Minor bug fixes and improved error messages.
- Add a LFW classification experiment and an outlier detection script.
- See this blog post for an overview and the GitHub Milestone for a high-level issue summary.
- Training improvements from resulting in an accuracy increase from 76.1% to 92.9%, which are from Bartosz Ludwiczuk's ideas and implementations in this mailing list thread. These improvements also reduce the training time from a week to a day.
- Nearly halved execution time thanks to Hervé Bredin's suggestions and sample code for image alignment in Issue 50.
- Hosted Python API Documentation.
- Docker automated build online.
- Initial automatic tests written in tests.
- Tests successfully passing in the Docker automated build in Travis.
- Add util/profile-pipeline.py to profile the overall execution time on a single image.
- Fix debug mode of NaiveDlib alignment.
- Add util/prune-dataset.py for dataset processing.
- Correct Docker dependencies.
- Initial release.