Progressive Growing of GANs implemented with chainer
python 3.5.2 + chainer 3.0.0
$ python3 ./train.py -g 0 --dir ./train_images/ --epoch 100 --depth 0
You can train models with ./train.py.
When depth = n, generated images are 2^{n+2} x 2^{n+2} size.
$ ./batch.sh 100
batch.sh automatically trains models gradually (through 4 x 4 to 256 x 256).
You should tune delta and epoch when it changes too quickly or too slowly.
$ python3 ./analogy.py --gen results/gen --depth 0
You can generate images with analogy.py
$ wget https://www.dropbox.com/s/dvnxb4vur6fasei/gen_yui_model
$ python3 ./analogy.py --gen gen_yui_model --depth 6
You can use the pre-trained model.
It generates 256 x 256 size Ichii Yui's images.
[1] http://research.nvidia.com/publication/2017-10_Progressive-Growing-of
The original paper
[2] https://github.com/dhgrs/chainer-WGAN-GP
WGAN-GP implemented with chainer.
[3] http://joisino.hatenablog.com/entry/2017/11/07/200000
My Blog post related to this repository.