memory format warning when using DiffAugment

[STomoya]

The warning bellow is produced when using Differentiable Augmentation(DiffAugment).

[W TensorIterator.cpp:924] Warning: Mixed memory format inputs detected while calling the operator. The operator will output channels_last tensor even if some of the inputs are not in channels_last format. (function operator())

I am guessing this only happens in PyTorch ver1.6.0 or higher.
PR#37968@pytorch/pytorch

DiffAugment is an augmentation method, therefore I believe this will not affect the results much.

[STomoya]

Notifications about kornia ver. DiffAugment

pros

• no warning

cons

• using nightly version of kornia.
• when augmenting colors in kornia, the values are clipped to fit between [0, 1], but the original implementation does not clip values.
  I don't believe this will cause a huge problem, but use the original one with PyTorch < 1.6.0 if it makes you nervous.

[STomoya]

Also it shouldn't be used to StyleGAN, because the original implementation doesn't use tanh before the output of the generator for each resolution. (not in my case though...)