Generative Adversarial Networks for Image Super-Resolution: A Survey

TL;DR

This survey addresses the problem of single image super-resolution (SISR) by organizing and evaluating generative adversarial networks (GANs) across supervised, semi-supervised, and unsupervised settings. It analyzes how GAN architectures, priors, loss functions, and multi-task strategies influence SR quality and robustness, and compares methods on public datasets using metrics such as $PSNR$ and $SSIM$, alongside qualitative assessments. The review highlights major challenges—training instability, high computational cost, and generalization to real-world degradations—and proposes future directions including lightweight designs, self-supervised pretraining, and multi-task frameworks to improve practical impact. By compiling 209 papers and providing structured taxonomies and benchmarks, the work serves as a comprehensive reference for researchers and practitioners navigating GAN-based SR methods.

Abstract

Single image super-resolution (SISR) has played an important role in the field of image processing. Recent generative adversarial networks (GANs) can achieve excellent results on low-resolution images. However, there are little literatures summarizing different GANs in SISR. In this paper, we conduct a comparative study of GANs from different perspectives. We begin by surveying the development of GANs and popular GAN variants for image-related applications, and then analyze motivations, implementations and differences of GANs based optimization methods and discriminative learning for image super-resolution in terms of supervised, semi-supervised and unsupervised manners, where these GANs are analyzed via integrating different network architectures, prior knowledge, loss functions and multiple tasks. Secondly, we compare the performances of these popular GANs on public datasets via quantitative and qualitative analysis in SISR. Finally, we highlight challenges of GANs and potential research points for SISR.

Figures (10)

• Figure 1: Outline of this overview. It mainly consists of basic frameworks, categories (i.e., supervised, semi-supervised and unsupervised GANs), performance comparison, challenges and potential directions.
• Figure 2: Architecture of generative adversarial network (GAN).
• Figure 3: Architecture of bidirectional generative adversarial network (BiGAN).
• Figure 4: Architecture of cycle-consistent adversarial network (CycleGAN).
• Figure 5: Architecture of StyleGAN.