Open shaoqb opened 3 years ago
https://github.com/facebookresearch/unbiased-teacher/blob/main/ubteacher/engine/trainer.py#L286
It seems that your teacher model does not freeze model weight and convert to eval mode?
I asked a similar question here the other day #47
Do you compare the experimental results of eval mode and training mode? I am experimenting on my data set, and there is probably a 2 point difference in results
https://github.com/facebookresearch/unbiased-teacher/blob/main/ubteacher/engine/trainer.py#L286
It seems that your teacher model does not freeze model weight and convert to eval mode?