sud0301 / semisup-semseg

136 stars 30 forks source link

About cosine_loss in MLMT #16

Open HHuiwen opened 3 years ago

HHuiwen commented 3 years ago

Hi authors, Thank you very much for sharing the code! In the paper, consistency loss is used for all available samples, but Why you only compute unlabeled data in the code? Is there any wrong with train_mlmt.py?

image

image