stylegan2-pytorch

stylegan2-pytorch copied to clipboard

Would it be possible to add an option to generate an augmented dataset using the --aug-prob, --dataset-aug-prob and --aug-types settings?

It's unclear to me how the different augment types and probabilities could impact my data, and it'd be informative to be able to take my dataset, augment it, and output the augmented dataset. This augmented dataset could be inspected for possible issues with augmentation and items could be excluded, then the model could be trained on the edited augmented dataset.

THIS! For instance, my dataset was quite negatively affected by the cropping/cutout augmentation, and it took me a while to realize the issue. Previewing the augmentation would be very cool.