Reducing distortions in Real World Image Super Resolution using Attention

Seungjun Jung, Changick Kim

Keywords: super resolution, deep learning

Abstract: In the real-world super-resolution (SR), we focus on enhancing perceptual quality beyond conventional resolution enhancement, aiming to address real-world degradations using Generative Adversarial Networks (GANs). GANs hold the potential for restoring fine details in low-resolution images, yet their generative nature may introduce distortions. Our contribution involves a novel approach that improves perceptual quality while minimizing distortions in real-world SR images through the strategic use of residual connections and an attention map. This approach has a simple structure and can be used to improve the performance of previously published SR models by using them as a backbone. We show in our experiments that our proposed method successfully reduces the distortions derived from GANs, thus improving the perceptual quality.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6048