FedSR: Frequency-Aware Enhancement for Diffusion-based Image Super-Resolution

Yueying Li, Hanbin Zhao, Jiaqing Zhou, Guozhi Xu, Tianlei Hu, Gang Chen, Haobo Wang

Keywords: Image Super-resolution, Frequency-Domain, Diffusion Models

TL;DR: A frequency-aware enhancement framework for diffusion-based image super-resolution

Abstract: Image super-resolution (ISR) is a classic and challenging problem in low-level vision because the data collection process often introduces complex and unknown degradation patterns. Leveraging powerful generative priors, diffusion-based algorithms have recently established new state-of-the-art ISR performance. Despite the promise, current diffusion-based ISR methods mostly focus on the spatial domain. To bridge this gap, we first experimentally validate that the key to solving the ISR problem lies in addressing the degradation of image amplitude information and high-frequency details. Based on this, we propose a novel $\textit{training-free}$ frequency-aware enhancement framework ($\textbf{FedSR}$) for diffusion-based ISR methods, which consists of two critical components. Firstly, we design the Amplitude Enhancement Module (AEM), which selectively enhances crucial amplitude channels through weighted optimization. Secondly, we introduce the High-Frequency Enhancement Module (HEM) that adaptively masks the skip features to perform high-pass filtering. Through extensive evaluations on both synthetic datasets and real-world image collections, our method demonstrates outstanding performance in reproducing realistic image details without additional tuning. For instance, FedSR improves StableSR across three datasets by $\textbf{+10.53}\%$ on MUSIQ metric.

Primary Area: applications to computer vision, audio, language, and other modalities

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/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 3824