hfslyc
commented
4 years ago It's because we use the imagenet pretrained model for initialization. The pretrained models work best if the same normalization is adopted during further training process.
Closed bitwalt closed 4 years ago
hfslyc
commented
4 years ago It's because we use the imagenet pretrained model for initialization. The pretrained models work best if the same normalization is adopted during further training process.
bitwalt
commented
4 years ago Thank you! :smile:
Why datasets are not normalized with own mean and stdv?