carpedm20 / BEGAN-tensorflow

Tensorflow implementation of "BEGAN: Boundary Equilibrium Generative Adversarial Networks"

BEGAN in Tensorflow

Tensorflow implementation of BEGAN: Boundary Equilibrium Generative Adversarial Networks.

Requirements

• Python 2.7 or 3.x
• Pillow
• tqdm
• requests (Only used for downloading CelebA dataset)
• TensorFlow 1.3.0

Usage

First download CelebA datasets with:

$ apt-get install p7zip-full # ubuntu
$ brew install p7zip # Mac
$ python download.py

or you can use your own dataset by placing images like:

data
└── YOUR_DATASET_NAME
    ├── xxx.jpg (name doesn't matter)
    ├── yyy.jpg
    └── ...

To train a model:

$ python main.py --dataset=CelebA --use_gpu=True
$ python main.py --dataset=YOUR_DATASET_NAME --use_gpu=True

To test a model (use your load_path):

$ python main.py --dataset=CelebA --load_path=CelebA_0405_124806 --use_gpu=True --is_train=False --split valid

Results

Generator output (64x64) with gamma=0.5 after 300k steps

Generator output (128x128) with gamma=0.5 after 200k steps

Interpolation of Generator output (64x64) with gamma=0.5 after 300k steps

Interpolation of Generator output (128x128) with gamma=0.5 after 200k steps

Interpolation of Discriminator output of real images

Related works

• DCGAN-tensorflow
• DiscoGAN-pytorch
• simulated-unsupervised-tensorflow

Author

Taehoon Kim / @carpedm20