Running ROI detection at training
Opened this issue · 0 comments
antkak commented
Is there a reason why we run ROI detection for training and keep those with large overlap, instead of just using the ground truth bounding boxes with transitional + scaling (random) augmentation?