Boundary-seeking generative adversarial networks (BGAN)

as featured in the paper: https://arxiv.org/abs/1702.08431v2

Requirements (rough estimate)

Basic instructions

Note: Very basic. In-depth instuctions forthcoming.

Datasets are available via Fuel: http://fuel.readthedocs.io/en/latest/built_in_datasets.html

Install MNIST:

$ cd <Dataset directory>

$ fuel-download binarized_mnist

$ fuel-convert binarized_mnist

Install CelebA:

$ cd <Dataset directory>

$ fuel-download celeba

$ fuel-convert celeba 64

Usage

For simple BGAN running on discrete MNIST:

python main_discrete.py -o <Output directory -S <Path to MNIST hdf5>

For simple BGAN running on continuous CelebA:

python main_continuous.py -o <Output directory> -S <Path to CelebA hdf5>

Basic documentation found in:

python main_continuous.py --help

Note: Published versions of the model are available in the code, and instructions to reproduce will be added soon.

If there are bugs or clarity is needed to run models, please add to the Issues.