search

lucidrains / lightweight-gan

Implementation of 'lightweight' GAN, proposed in ICLR 2021, in Pytorch. High resolution image generations that can be trained within a day or two
MIT License
1.63k stars 222 forks source link

Is it possible to add the experimental options from your StyleGAN2 to this GAN model? #57

Closed zxsitu closed 3 years ago

zxsitu commented 3 years ago

Hello, this GAN you provided is very meaningful for my research as I have a limited set of images (only close to 9,000). I would like to ask if you have the possibility to add your experimental features in StyleGAN2 （https://github.com/lucidrains/stylegan2-pytorch#experimental） to the training of this GAN model? This could enrich my research. Your experimental options in StyleGAN2 are these. 1、Top-k Training for Generator 2、Feature Quantization 3、Contrastive Loss Regularization 4、Relativistic Discriminator Loss

jonas-klesen commented 3 years ago

I would also love to see this. Especially top-k training and a parameter to increase network capacity.

Mut1nyJD commented 3 years ago

@deklesen

I would also love to see this. Especially top-k training and a parameter to increase network capacity.

It already does top-k training explicitly for the generator if you look closely in the training loop you can see it does.