Train a state-of-the-art spectral normalization GAN with AdaBelief https://github.com/juntang-zhuang/Adabelief-Optimizer

This repo is forked from PyTorch-StudioGAN github repository, with the only difference in optimizer.

pip install adabelief-pytorch==0.0.5

For other dependencies, see https://github.com/POSTECH-CVLab/PyTorch-StudioGAN/blob/master/environment.yml

python main.py -t -e -c configs/CIFAR10/SNGAN-adabelief.json

You can modify parameters epsilon and rectify in load_framwork.py https://github.com/juntang-zhuang/SNGAN-AdaBelief/blob/master/load_framework.py#L135

Results for Adam is directly taken from the training log of official implementation https://github.com/POSTECH-CVLab/PyTorch-StudioGAN/blob/c98d0e94d98a97e14165bd42bd1416027cf78f4d/logs/CIFAR10/SNGAN-train-2020_09_18_14_37_00.log