search

dvschultz / stylegan2-ada-pytorch

StyleGAN2-ADA - Official PyTorch implementation
https://arxiv.org/abs/2006.06676
Other
245 stars 121 forks source link

[Request] Add support for aesthetic score -aided dataset into the discriminator and loss functions. #41

Open nom57 opened 2 years ago

nom57 commented 2 years ago

Add aesthetic-rating aided dataset support into the discriminator loss function. Add a function that could penalize the discriminator for thinking an image from the dataset is "more real" than one which has a higher aesthetic-score rating (could be an aesthetic score system beetwen 1 to 5 or 1 to 10). We could organize a dataset based on aesthetic scores by seperating them into seperate folders (example : folders named 1,2,3,4,5,6,7,8,9,10 , for 1-10 aesthetic score rating).

This would greatly help use-cases where you have a lot of images for a dataset but only a select few are good enough in quality and creativity, but you still wanna keep the diversity of a much larger dataset.

dvschultz commented 2 years ago

im not really updating this repo anymore beyond managing dependencies. If anyone wants to work on this you’re welcome to and submit a PR.