SRGAN-with-WGAN-Loss-TensorFlow

SRGAN with WGAN loss function in TensorFlow

Introduction

This code mainly address the problem of super resolution, Super Resolution Generative Adversarial Networks

There are four different from the paper:

1. The loss function, we use WGAN loss, instead of standard GAN loss.
2. The network architecture, Because of our poor device, in generator, we just use 5 residual block (paper: 16), and in discriminator, we use the standard DCGAN’s discriminator.
3. The training set, device problem again, we just use a part of ImageNet (ImageNet Val) which just contains 50,000 images.
4. The max iteration, we just train the model about 100,000 iterations, instead of the paper 600,000.

How to use

1. Download the dataset ImageNet Val
2. unzip dataset and put it into the folder ‘ImageNet’

   ├── test
   ├── save_para
   ├── results
   ├── vgg_para
   ├── ImageNet
     ├── ILSVRC2012_val_00000001.JPEG
     ├── ILSVRC2012_val_00000002.JPEG
     ├── ILSVRC2012_val_00000003.JPEG
     ├── ILSVRC2012_val_00000004.JPEG
     ├── ILSVRC2012_val_00000005.JPEG
     ├── ILSVRC2012_val_00000006.JPEG
     ...

3. execute the file main.py

   Requirements

• python3.5
• tensorflow1.4.0
• pillow
• numpy
• scipy
• skimage

  Results

  Train procedure WGAN Loss

  

Down sampled	Bicubic (x4)	SRGAN (x4)

Reference

[1] Ledig C, Theis L, Huszár F, et al. Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network[C]//CVPR. 2017, 2(3): 4.