In the fall semester, we focused on improving the system’s overall performance. We upgraded our traffic sign detectors by using focal loss, which handled the issue of unbalanced data and improved the sign detectors’ performances on hard cases. Besides, we generated synthetic data for the infrequent categories to further improve the detector’s performance on those hard categories. Finally, we introduced some new steps in the post-processing parts to get more consistent traffic sign projections.
2020 Fall Presentation Slides – link
2020 Fall Presentation Video – link