[ICLR 2022] Official pytorch implementation of "Uncertainty Modeling for Out-of-Distribution Generalization" in International Conference on Learning Representations (ICLR) 2022.
141
stars
16
forks
source link
Question about FrozenBatchNorm in Semantic Segmentation #7
I have a question about the FrozenBatchNorm in semantic segmentation. I have read about the code of FrozenBatchNorm2d you provided. It seems that the FrozenBatchNorm2d is an identity operation. Which means that there is no normalization operation during training. Why you use FrozenBatchNorm during training in semantic segmentation?
Hi, thanks for your question.
The implementation is from the original semantic segmentation FADA repo[https://github.com/JDAI-CV/FADA], which we do not change for a fair comparison.
Hi, author. Thanks for open-sourcing the code.
I have a question about the FrozenBatchNorm in semantic segmentation. I have read about the code of FrozenBatchNorm2d you provided. It seems that the FrozenBatchNorm2d is an identity operation. Which means that there is no normalization operation during training. Why you use FrozenBatchNorm during training in semantic segmentation?