Contrast difference in Autoencoder output images

vision
1

I trained an Autoencoder network for images. Input image transform is :

transforms.Compose([transforms.ToPILImage(),transforms.ToTensor(),
										transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

I used tanh in final layer and I am using torchvision.utils.save_image with normalize=True, but I observe a contrast difference between original image and generated one. What might be the problem here?