Residual-Dense-Network-Trained-with-cGAN-for-Super-Resolution

This repository is as a research project in the field of super resolution. It uses RDN as the generator and spectral norm is used in discriminator.

Introduction

This is a trial for super-resolution

The residual dense network has many advantages for reconstructing SR images, and we use GANs to enhance RDN. The core idea is from the following two papers:

1. Residual Dense Network for Image Super-Resolution
2. cGANs with projection discriminator

Generator: Residual Dense Network

Discriminator: cGAN projection

Results

These results is just trained about 200,000 iterations (full: 600,000) with batch size of 16.

Raw	Bicubic(x4)	RDN_GAN(x4)

Reference

[1] Zhang Y, Tian Y, Kong Y, et al. Residual dense network for image super-resolution[C]//The IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018.

[2] Miyato T, Koyama M. cGANs with projection discriminator[J]. arXiv preprint arXiv:1802.05637, 2018.