PyTorch Torchvision UnNormalize (reverse Normalize)

unnormalize.py

import torchvision

class UnNormalize(torchvision.transforms.Normalize):
    def __init__(self,mean,std,*args,**kwargs):
        new_mean = [-m/s for m,s in zip(mean,std)]
        new_std = [1/s for s in std]
        super().__init__(new_mean, new_std, *args, **kwargs)
        
# imagenet_norm = dict(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])
# UnNormalize(**imagenet_norm)