code for steinGAN - Learning to Draw Samples: With Application to Amortized MLE for Generative Adversarial Learning
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.

README.md

SteinGAN

This repository contains the code for the paper Learning to Draw Samples: With Application to Amortized MLE for Generative Adversarial Learning.

The code is based on the excellent implementation of DCGAN.

MNIST

We evaluate our methods on MNIST and use a simple Gaussian-Bernoulli Restricted Boltzmann Machines (RBM) as our energy-based model, which allows us to accurately evaluate the test likelihood. SteinGAN without the kernel tends to produce much less diverse images. This suggests that the repulsive term is responsible for generating diverse images in SteinGAN.

We use the implementation of Wu et al., 2017 to evaluate the test log-likelihood.

SteinGAN SteinGAN without kernel
TestLL: -771.2 Test LL: -989.6
-------- --------------------

CIFAR-10

Results on CIFAR-10. For more details, please refer to our paper.

CelebA 64x64

CelebA images generated by SteinGAN.

LSUN 64x64

Citation

If you find SteinGAN helpful for your research, please cite the following papers:

  • Dilin Wang and Qiang Liu. Learning to Draw Samples: With Application to Amortized MLE for Generative Adversarial Learning. arXiv preprint arXiv:1611.01722, 2016.
  • Alec Radford, Luke Metz, Soumith Chintala. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. arXiv preprint arXiv:1511.06434. 2015.

Feedback

Feedback is greatly appreciated. If you have any questions, comments, issues or anything else really, shoot us an email.

All rights reserved.