- 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
Python API Documentation.
- Docker automated build online.
- Initial automatic tests written in tests.
- Tests successfully passing
in the Docker automated build in Travis.
to profile the overall execution time on a single image.