d-li14/mobilenetv2.pytorch

Question about normalization in data preprocessing

zhangyongshun opened this issue · 0 comments

Hi,
Thanks for your wonderful work!
I find the normalization in your codes does not use ``transforms.ToTensor(), transforms.Normalize()'' provided by PyTorch. The normalization you used is to sub and div on [0, 255] space directly.
I wonder why you use this normalization, dose it make big difference to the results?