search

wuhuikai / FastFCN

FastFCN: Rethinking Dilated Convolution in the Backbone for Semantic Segmentation.
http://wuhuikai.me/FastFCNProject
Other
838 stars 148 forks source link

about syncbn #97

Closed xLuge closed 3 years ago

xLuge commented 3 years ago

hi, why use syncbn in training and then use BatchNorm in testing ? should we freeze syncbn in testing or finetune ? and how ?

wuhuikai commented 3 years ago

syncbn is for large batchsize in training. just freeze it in testing phase.

xLuge commented 3 years ago

so,the way to freeze syncbn is setting norm_layer = BatchNorm in test.py? just like this ![image](https://user-images.githubusercontent.com/37794367/117934230-098ff500-b335-11eb-94ed-b65d98fb947d.png

wuhuikai commented 3 years ago

After you call model.eval(), BN is freezed.