Semi-supervised-learning icon indicating copy to clipboard operation
Semi-supervised-learning copied to clipboard

Published 20 hours ago • verified publisher icon microsoft

Questions about batch normalization handling.

Open pm25 opened this issue 1 year ago • 0 comments

I've noticed that in certain methods like PiModel and MeanTeacher, batch normalization is frozen during inference on unlabeled data. I'm curious about the rationale behind this decision. Unlike in FixMatch, where this operation isn't performed. Could you please shed some light on this? Thank you!

https://github.com/microsoft/Semi-supervised-learning/blob/6dc008347b8ed2a546d788dcd01a476fdced05a2/semilearn/algorithms/pimodel/pimodel.py#L43-L51

https://github.com/microsoft/Semi-supervised-learning/blob/6dc008347b8ed2a546d788dcd01a476fdced05a2/semilearn/algorithms/meanteacher/meanteacher.py#L44-L57

pm25 avatar Mar 04 '24 16:03 pm25