Implementation for MAGAN: Margin Adaptation for Adversarial Networks

Tensorflow implementation of MAGAN: Margin Adaptation for Generative Adversarial Networks.

Requirements

• Python 2.7
• Pillow
• TensorFlow 1.0.1
• PrettyTensor

Usage

First download CelebA dataset at link.
and/or CIFAR-10 (python version) at link:

To train a model:

$ python gan_release.py [cifar/celebA] {dataset_dir}
$ python gan_release.py mnist #data_dir not needed for mnist 

Results

Generator output on MNIST, CIFAR-10, and CelebA

MNIST output

CIFAR-10 output

CelebA output

Author

Ruohan Wang