Collaborative Sampling in Generative Adversarial Networks

This repository provides a basic PyTorch implementation of the Collaborative Sampling in Generative Adversarial Networks.

Paper | Poster | TensorFlow

Requirements

To install requirements:

pip install -r requirements.txt

Training

To train the GANs, run this command:

python main.py --mode="train" --niter=10000

Sampling

To collaboratively sample from the trained GANs, run this command:

python main.py --mode="collab" --ckpt_num=3000 --niter=3000 --lrd=5e-2

Result

	Real	Vanilla Sampling GAN 1K Iter	Vanilla Sampling GAN 3K Iter	Vanilla Sampling GAN 10K Iter	Collab Sampling GAN 3K Iter
Samples
KDE

Citation

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

@inproceedings{liu2019collaborative,
  title={Collaborative Sampling in Generative Adversarial Networks},
  author={Liu, Yuejiang and Kothari, Parth and Alahi, Alexandre},
  booktitle={Thirty-first AAAI conference on artificial intelligence},
  year={2020}
}