ToTensor already “normalizes” the [0,255] uint input image into [0.0, 1.0] floats. Hence, the mean and std args expected by transforms.Normalize should be also scaled down to the [0.0, 1.0] range.
1 Like