Animal Faces-HQ v2 (AFHQ) - Debias Estimation Loss Comparison Test

This uses the Animal Faces-HQ v2 dataset by Clova AI Research.

Might be neat for a training base, or in a merge. I have another one which includes FFHQ 512x512. It seems to do these same animals better, but also do faces similar to stylegan/thispersondoesnotexist.

Only the unet was trained. There are two versions; one that uses Debias Estimation Loss, and another which doesn't.

This could likely look much better if I did more epoch than one epoch, or fiddled with the training settings. I may come back to this an run a few more epochs since it only took ~1.5 hours to train.

@inproceedings{choi2020starganv2,
  title={StarGAN v2: Diverse Image Synthesis for Multiple Domains},
  author={Yunjey Choi and Youngjung Uh and Jaejun Yoo and Jung-Woo Ha},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}