EntropicGANs_meet_VAEs

Official implementation of "EntropicGANs meet VAEs: A Statistical Approach to Compute Sample Likelihoods in GANs"

Training

To train the model, run

python src/main.py --savename mnist --savedir results/MNIST --gan_mode swgan --usePrimalLoss

To compute sample likelihoods on the trained EntropicGAN model, run

python src/main.py --savename mnist --savedir results/MNIST --loadpath results/MNIST/models/model_5000.ckpt --gan_mode swgan --mode eval --evalroot 'path to dataset whose likelihood we wish to compute'

The likelihood scores for samples are stored as a numpy array.

Citation

If you use this code for your research, please cite

@article{Balaji2018Entropic,
author    = {Yogesh Balaji and
             Hamed Hassani and
             Rama Chellappa and
             Soheil Feizi},
title     = {Entropic GANs meet VAEs: {A} Statistical Approach to Compute Sample
             Likelihoods in GANs},
journal   = {CoRR},
volume    = {abs/1810.04147},
year      = {2018},
url       = {http://arxiv.org/abs/1810.04147},
}