danielyou0230 / SRGAN

Super-Resolution Generative Adversarial Networks Implementation

Super-Resolution Generative Adversarial Networks (SRGAN)

About this project

This project aims to implement SRGAN based on the Christian Ledig et al's "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network" (See: https://arxiv.org/abs/1609.04802) with Tensorflow.

In this project, we might incorporate the contents in Martin Arjovsky et al's "Wasserstein GAN" (See: https://arxiv.org/abs/1701.07875) and Huikai Wu el al's "GP-GAN_Towards Realistic High-Resolution Image Blending" (See: https://arxiv.org/abs/1703.07195) to achieve better performance or so.

Current Progress

Implementing VGG19 (pending for verification)
Implementing SRGAN (revising loss function from https://github.com/tadax/srgan )

Project Roadmap

1. Implement VGG19
2. Verify VGG19
3. Implement SRGAN
4. Verify SRGAN
5. Train VGG19
6. Train SRGAN
7. Evaluate performance on SRGAN

Issues

#1. How to train VGG19 with Super-Resolution purpose?
#2. Implementation on training SRGAN

References

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
Christian Ledig et al
https://arxiv.org/abs/1609.04802

Wasserstein GAN
Martin Arjovsky, Soumith Chintala, Léon Bottou
https://arxiv.org/abs/1701.07875

GP-GAN: Towards Realistic High-Resolution Image Blending
Huikai Wu, Shuai Zheng, Junge Zhang, Kaiqi Huang
https://arxiv.org/abs/1703.07195

How to Train Your DRAGAN
Naveen Kodali, Jacob Abernethy, James Hays, Zsolt Kira
https://arxiv.org/abs/1705.07215

Deep Residual Learning for Image Recognition
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
https://arxiv.org/abs/1512.03385