search

bryandlee / animegan2-pytorch

PyTorch implementation of AnimeGANv2
MIT License
4.41k stars 641 forks source link

Training details #37

Open chrisha97 opened 2 years ago

chrisha97 commented 2 years ago

Hi! Great work! May I ask what GAN loss are you using? I'm trying to achieve a similar effect with WGAN, but the result is quite poor. Also, you mentioned that you're using a face segmentation model to separate background from the face. May I ask you which model are you using?

bryandlee commented 2 years ago

Hi, I used non-saturating loss for the GAN, and used deeplabv3 for the face part segmentation. If background mask is all you need, then u2net gives finer results.