Experiment on range of percentages of negative samples for training #79
Description
Started with 5% of dataset being negative frames. This incurs a model that still has way too many false positives. Should increase number of negative frames until there is a detriment in mAP score.
- 5% negative frames
- 10% negative frames
- 20% negative frames