More questions about ROI heads and pseudo-label generation

[vadimkantorov]

1. In what part of code do you handle this? In practice, conflicts happen when we force the yˆ(·, r) to be a one-hot vector since the same region can be chosen to be positive for different ground-truth classes, especially in the early stages of training. Our solution is to use that class for pseudo-label rˆ which has a higher predicted score s(c, rˆ).

2. What scores are used for generating supervision for student branches? It seems to me that you normalize scores across classes. Is it true?

   wetectron/wetectron/modeling/roi_heads/weak_head/loss.py

   Line 257 in 44e6fa9

   source_score = final_score_per_im if i == 0 else F.softmax(ref_scores[i-1][idx], dim=1)

Thanks!