Why stylegan images' channel statistics are random?

[zjysteven]

Hi,

Thanks for making the code available. It helps. I was wondering why in the normalization of stylegan's outputs the channel mean and std are randomly generated (as below). Also, does this normalization matter a lot? What if one just uses the min-max normalization?

https://github.com/mbaradad/learning_with_noise/blob/8b5aa7f792b69fcb7a331da03cf3ecbcca8b08cd/generate_datasets/stylegan/generate_dataset.py#L61-L66

Thanks!