Stacked Generative Adversarial Networks
This repository contains code for the paper "Stacked Generative Adversarial Networks", CVPR 2017. Part of the code is modified from OpenAI's implementation of Improved GAN.
Architecture
Samples
Performance Comprison on CIFAR-10
| Method    | Inception Score |
|---|---|
| Infusion training  |  4.62 ± 0.06   |
| GMAN (best variant) |  5.34 ± 0.05 |
| LR-GAN |  6.11 ± 0.06 |
| EGAN-Ent-VI |  7.07 ± 0.10 |
| Denoising feature matching |  7.72 ± 0.13 |
| DCGAN | Â 6.58 |
| SteinGAN | Â 6.35 |
| Improved GAN（best variant) |  8.09 ± 0.07 |
| AC-GAN |  8.25 ± 0.07 |
| SGAN (ours)  |  8.59 ± 0.12 |
Citations
If you use the code in this repository in your paper, please consider citing:
@inproceedings{huang2017sgan,
title={Stacked Generative Adversarial Networks},
author={Huang, Xun and Li, Yixuan and Poursaeed, Omid and Hopcroft, John and Belongie, Serge},
booktitle={CVPR},
year={2017}
}
Contact
If you have any questions about the code, feel free to email me ([email protected]).