figure2
Closed this issue · 2 comments
a-drop-in-the-ocean456 commented
zimoqingfeng commented
During training, we generate logs for recording "the number of total valid samples" and "the number of positive samples" in each level, and then draw a box polt according to their proportions.
More details in
https://github.com/zimoqingfeng/UMOP/blob/master/mmdet/models/dense_heads/anchor_head.py#L417
https://github.com/zimoqingfeng/UMOP/blob/master/mmdet/models/dense_heads/fcos_head.py#L229
https://github.com/zimoqingfeng/UMOP/blob/master/mmdet/models/dense_heads/vfnet_head.py#L394
a-drop-in-the-ocean456 commented
thank you for your answer.