Generative Adversarial Networks for Image Super resolution A Survey is conducted by Chunwei Tian, Xuanyu Zhang, Jerry Chun-Wei Lin, Wangmeng Zuo and Yanning Zhang and it is available at https://arxiv.53yu.com/abs/2204.13620. It is reported by Cver and 52CV.

This paper is a complete summary of generative adversarial networks for image super-resolution, which is very meaningful to readers.

Absract

Single image super-resolution (SISR) has played an important role in the field of image processing. Recent generative adversarial networks (GANs) can obtain excellent results in terms of low-resolution images with small samples. However, there are little literatures to summarize different GANs in SISR. In this paper, we conduct comparative research of GANs, according to different perspectives. We first look through developments of GANs. Secondly, we give popular architectures about GANs in big samples and small samples for image applications. Then, we analyze motivations, implementations and differences of GANs based optimization method and discriminative learning for image super-resolution in terms of supervised, semi-supervised and unsupervised manners. Next, we compare performance of these popular GANs on public datasets via quantitative and qualitative analysis in SISR. Finally, we highlight challenges of GANs and potential research points on SISR.

Outline of this overview.

Architecture of generative adversarial network (GAN).

Architecture of bidirectional generative adversarial network (BiGAN).

Architecture of cycle-consistent adversarial network (CycleGAN).

Architecture of StyleGAN.

The structure of generator in the StyleGAN.

Introduction of many GANs.

Frame of popular GANs for image applications.

GANs on big samples for iamge generation.

GANs on big samples for object detection.

GANs on samll samples for style transfer.

GANs on samll samples for iamge inpainting.

Frame of GANs for image super-resolution.

Suervised GANs for image super-resolution in section IV part A para.1.

Suervised GANs for image super-resolution in section IV part A para.2 to section IV part A para.4.

Semi-suervised GANs for image super-resolution in section IV part B.

Unsuervised GANs based improved architectures for image super-resolution.

Unsuervised GANs based prior knowledge and improved loss functions for image super-resolution.

Unsuervised GANs based multi-tasks for image super-resolution.

Datasets (i.e., training datasets and test datasets) of GANs for image super-resolution.

Different GANs on image super-resolution for different training ways.

Environment configurations of different GANs for image-resolution.

PSNR and SSIM of different GANs via different training ways on Set14, BSD100 and DIV2K for image super-resolution.

Running time and parameters of different GANs for x4.

Visual images of different GANs on an image of BSD100 for ×4: (a) original image, (b) Bicubic, (c) ESRGAN, (d) RankSRGAN, (e) KernelGAN, and (f) PathSRGAN.

Visual images of different GANs on an image of Set14 for ×4: (a) original image, (b) Bicubic, (c) ESRGAN, (d) RankSRGAN, (e) KernelGAN, and (f) PathSRGAN.

Some popular super-resolution models based GANs are shown as follows.

Real-ESRGAN can be obtained at https://github.com/xinntao/Real-ESRGAN.

ESRGAN can be obtained at https://github.com/xinntao/ESRGAN.

SRGAN can be obtained at https://github.com/tensorlayer/srgan.

RankSRGAN can be obtained at https://github.com/WenlongZhang0517/RankSRGAN.

KernelGAN can be obtained at https://github.com/sefibk/KernelGAN.

If you cite this paper, please refer to the following formats:

1. Tian C, Zhang X, Lin J C W, et al. Generative Adversarial Networks for Image Super-Resolution: A Survey[J]. arXiv preprint arXiv:2204.13620, 2022.

2. @article{tian2022generative,

title={Generative Adversarial Networks for Image Super-Resolution: A Survey},

author={Tian, Chunwei and Zhang, Xuanyu and Lin, Jerry Chun-Wen and Zuo, Wangmeng and Zhang, Yanning},

journal={arXiv preprint arXiv:2204.13620},

year={2022}

}