about the pixel norm

[menggerSherry]

I got a problem, in the original paper the pixelwise feature vector normalization in generator is ：x/torch.rsqrt((x**2).mean(dim=1,keepdim=True)+epilon)

but in your code I see ：
x*torch.rsqrt((x**2).mean(dim=1,keepdim=True)+epilon)

Is it better to use your approach？