f-GANs in an Information Geometric Nutshell
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
models
LICENSE
README.md
fileUtil.py
kde.py
main.py
plot_results.py

README.md

f-GANs in an Information Geometric Nutshell

Pytorch implementation of f-GANs in an Information Geometric Nutshell

Prerequisites

Usage

Put both mnist and lsun in a folder called DATA_ROOT. Download lsun with https://github.com/fyu/lsun. MNIST will be downloaded automatically in the first run.

$ python download.py -o <DATA_ROOT> -c tower

Assume all experimental results are put in EXPERIMENTAL_RESULTS.

Evaluate a feedforward network with wasserstein GAN loss and mu-ReLU as the activation of hidden layers of the generator:

$ python main.py --dataset mnist --dataroot <DATA_ROOT> --cuda -D wgan -A mlp -H murelu --experiment <EXPERIMENTAL_RESULTS> --task mu

Evaluate DCGAN with GAN as the loss, and mu-ReLU as the activation of hidden layers of the generator:

$ python main.py --dataset lsun --subset tower --dataroot <DATA_ROOT> --cuda -D gan -A dcgan -H murelu --experiment <EXPERIMENTAL_RESULTS> --task mu

Author

Lizhen Qu / @qulizhen