chainer implementation of triple GANs
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md
make_mnist_datasets.py
make_mnist_npy.py
train_tripleGAN_chainer.py
utility.py

README.md

Sample code for Triple GAN

Discription

This software is a python implementation of Triple GAN using chainer.

See the paper. https://arxiv.org/abs/1703.02291

Notice that this code had not implement gaussian noise in discriminator, Because I don't know how to implement it in chainer.

Requirement

I only confirm the operation under this environment

  1. python 3.5.3
  2. chainer 3.3.0
  3. pillow 5.0.0
  4. numpy 1.14.0
  5. cupy 2.3.0

Usage

First of all, prepare mnist data as npy form.
python make_mnist_npy.py Then, you can get mnist_train_img.npy, mnist_train_label.npy, mnist_test_img.npy and mnist_test_label.npy.

To train this model, do python train_tripleGAN_chainer.py

Sample image

Sample image generated in generater after about 500 epochs. resultimage_580

Refference

I also made triple GAN code in tensorflow.
https://github.com/masataka46/tripleGAN
But, there are some bugs.

License

MIT