RankSRGAN: Generative Adversarial Networks With Ranker for Image Super-Resolution

By Wenlong Zhang, Yihao Liu, Chao Dong, Yu Qiao. IEEE International Conference on Computer Vision, 2019

409 citations recorded by the source corpus.

Generative Adversarial Networks (GAN) have demonstrated the potential to recover realistic details for single image super-resolution (SISR). To further improve the visual quality of super-resolved results, PIRM2018-SR Challenge employed perceptual metrics to assess the perceptual quality, such as PI, NIQE, and Ma. However, existing methods cannot directly optimize these indifferentiable perceptual metrics, which are shown to be highly correlated with human ratings. To address the problem, we propose Super-Resolution Generative Adversarial Networks with Ranker (RankSRGAN) to optimize generator in the direction of perceptual metrics. Specifically, we first train a Ranker which can learn the behavior of perceptual metrics and then introduce a novel rank-content loss to optimize the perceptual quality. The most appealing part is that the proposed method can combine the strengths of different SR methods to generate better results. Extensive experiments show that RankSRGAN achieves visually pleasing results and reaches state-of-the-art performance in perceptual metrics. Project page: https://wenlongzhang0724.github.io/Projects/RankSRGAN

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/1d01c7431458564aa64606b4b6a7f9161ee423b1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Chen Change Loy
3. Radu Timofte
4. Lei Zhang
5. Ming-Hsuan Yang
6. Yu Qiao
7. Chao Dong
8. Dacheng Tao

Citations in the corpus

• Efficient Image Super-Resolution Using Pixel Attention (2020, background): “e reconstruction branch and propose a U-PA block. Note that few studies have investigated attention schemes after upsampling. 2 Related Work 2.1 Ecient CNN for SR Recently, many deep neural networks [45,44,3,20,43,9] have been introduced to improve the reconstruction results. However, the huge amoun”
• PIPAL: a Large-Scale Image Quality Assessment Dataset for Perceptual Image Restoration (2020, methodology): “, ESRGAN [47] and RankSRGAN [61]) could still produce images with obvious unrealistic artifacts.”
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications (2020, neutral): “Other related methods can be found in [320]–[323].”
• FD-GAN: Generative Adversarial Networks with Fusion-discriminator for Single Image Dehazing (2020, background): “GAN has great potential to generate realistic images and have been introduced in various vision tasks, such as super-resolution (Zhang et al. 2019b), de-raining (Zhang et al. 2019a).”
• Frequency Separation for Real-World Super-Resolution (2019, methodology): “The recently introduced RankSRGAN [39] uses a method to train SR models on indifferentiable perceptual metrics.”
• Closed-Loop Matters: Dual Regression Networks for Single Image Super-Resolution (2020, methodology): “Based on DNNs, many methods have been proposed to improve SR performance [51, 26, 10, 12, 49].”
• Deep Unfolding Network for Image Super-Resolution (2020, result): “Apart from RCAN, IKC and IRCNN, we also include RankSRGAN [69] for comparison with our USR-GAN.”
• Learning Texture Transformer Network for Image Super-Resolution (2020, methodology): “Recent proposed RSRGAN [38] trained a ranker and used rank-content loss to optimize the perceptual quality, which achieved state-of-the-art visual results.”
• Activating More Pixels in Image Super-Resolution Transformer (2022, background): “To improve perceptual quality, [24,53,64,52] introduce adversarial learning to generate more realistic results.”
• Real-World Super-Resolution via Kernel Estimation and Noise Injection (2020, background): “Furthermore, some Generative Adversarial Networks (GAN)-based methods [21, 44, 39] pay more attention to visual effects, introducing adversarial losses and perceptual losses.”
• Exploring CLIP for Assessing the Look and Feel of Images (2022, background): “…interest to explore the possibility of a universal understanding of both look and feel , as it can not only save efforts on manual labeling but also facilitate the development of vision tasks such as restoration (Zhang et al. 2019; Jo, Yang, and Kim 2020; Wenlong et al. 2021; Hui et al. 2021).”
• Hierarchical Conditional Flow: A Unified Framework for Image Super-Resolution and Image Rescaling (2021, methodology): “For general image SR (×4), we compare HCFlow with state-of-the-art CNN-based and flowbased SR models, including the PSNR-oriented EDSR [29] and RRDB [43], perception-oriented ESRGAN [43] and RankSRGAN [53], as well as SRFlow [32].”
• SRFlow: Learning the Super-Resolution Space with Normalizing Flow (2020, methodology): “he input LR image. 5.2 General Super-Resolution Next, we evaluate our SRFlow for general SR on the DIV2K validation set. We compare SRFlow to bicubic, EDSR [24], RRDB [47], ESRGAN [47], and RankSRGAN [56]. Except for EDSR, which used DIV2K, all methods including SRFlow are trained on the train split”
• SRDiff: Single Image Super-Resolution with Diffusion Probabilistic Models (2021, neutral): “, 2017], RRDB, ESRGAN, RankSRGAN [Zhang et al., 2019] and SRFlow (τ = 0.”