Haochen-Wang409 / U2PL

[CVPR'22] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels
Apache License 2.0
426 stars 59 forks source link

Question about the teacher model #149

Closed Hugo-cell111 closed 11 months ago

Hugo-cell111 commented 1 year ago

Hi! Recently I have also tried to use student-teacher model in semi-supervised learning, and I have also met the same problem as you explained in this issue. But I wonder whether it is necessary to set the teacher model as model_t = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model_t)? Thanks!

Haochen-Wang409 commented 1 year ago

I think it could be necessary since the parameters of BN will be updated during the teacher forward by switching model_teacher.train().