Differences between the URCP paper and code

[hiker]

SSL4MIS/code/train_uncertainty_rectified_pyramid_consistency_2D.py

Line 165 in ad16872

 torch.log(outputs_soft[args.labeled_bs:]), preds[args.labeled_bs:]), dim=1, keepdim=True) 

I noticed a log operation employed before calculating KL divergence. I was confused about it, and it is different from Eq. (4) in either the MIA paper or the MICCAI paper.

Besides, I'm not sure whether the subscripts of Eq. (4) in both paper are wrong, because $P_C$ is not mentioned again, and $P_{avg}$ is not appear in the formulas.