A pytorch implementation for OGAN 《Orthogonal Generative Adversarial Network》
python train.py --dataset_path <img_dir> --batch_size 16 --epochs 50
The checkpoints will be saved in dir checkpoints
, and the sample images in dir output
while training.
Replace the default checkpoints path in gen_imgs.py
, and run:
python gen_imgs.py
Using celeba dataset, after 3 epochs (About 2 hours):