YU1ut/MixMatch-pytorch

It seems the mean teacher does not contribute to the loss.

jonkoi opened this issue · 2 comments

The teacher model ema_model is updated every step, but it is never use for forward prop. Thus, I am not sure if mean teacher does anything.

The ema_model only used in valuation and test step. Actually, I think it does not work to improve the model's efficiency. @jonkoi

I see. Thanks for clearing this up for me. Closing this issue.