GlassyWing / ogan-torch

An unofficial pytorch implementation for OGAN

ogan-pytorch

A pytorch implementation for OGAN 《Orthogonal Generative Adversarial Network》

Usage

Training

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.

Generation

Replace the default checkpoints path in gen_imgs.py, and run:

python gen_imgs.py 

Example

Using celeba dataset, after 3 epochs (About 2 hours):

Generation