EvolutionaryGAN

We provided Theano implementations for Evolutionary Generative Adversarial Networks (E-GAN). Meanwhile, we are working on new Pytorch-based implementations.

Getting started

• Clone this repo:

git clone git@github.com:WANG-Chaoyue/EvolutionaryGAN.git
cd EvolutionaryGAN

• Install Theano 1.0.0+, lassagne 0.2+ and other dependencies (requirements.txt).

Datasets

The proposed E-GAN was trained on two synthesis dataset and three real-world datasets. Among them, the two mixture Gaussians datasets are adopted from here. For three real-world datasets, you should download them first from Cifar-10, LSUN bedroom, and CelebA, and then moving them into datasets. Then,

cd dataset
python dataset.py

The download CelebA or LSUN bedroom datasets can be converted to 'hdf5' files.

Training

• Train a model (take cifar-10 as an example)

cd cifar10
python train_cifar10.py

Note that related hpyer-parameters can be configured within 'train_cifar10.py'

vim train_cifar10.py

Citation

If you use this code for your research, please cite our paper.

@article{wang2018evolutionary,
  title={Evolutionary Generative Adversarial Networks},
  author={Wang, Chaoyue and Xu, Chang and Yao, Xin and Tao, Dacheng},
  journal={arXiv preprint arXiv:1803.00657},
  year={2018}
}

Related links

Evolving Generative Adversarial Networks | Two Minute Papers #242