PyTorch Implementation of EigenGAN
Train
python train.py [image_folder_path] --name [experiment name]
Test
python test.py [ckpt path] --traverse
FFHQ
samples (no truncation)
Learned subspace: L0 D1
Learned subspace: L1 D2
Anime
samples (no truncation)
Learned subspace: L0 D5
Learned subspace: L1 D2
Note
Default training configurations are different from the original implementation
Tested on python 3.8 + torch 1.8.1
Issue
Some of the subspace layers seem to collapse (does not contribute to the output) as the training proceeds and FID gets better.