Fourier-Guided Attention Upsampling for Image Super-Resolution

TL;DR

This paper introduces Frequency-Guided Attention (FGA), a lightweight upsampling module that improves high-frequency detail reconstruction and spectral fidelity in image super-resolution, outperforming traditional methods with minimal additional parameters.

Contribution

FGA integrates Fourier-based encoding, adaptive spatial alignment, and spectral supervision, providing a novel, scalable upsampling approach that enhances detail preservation in super-resolution tasks.

Findings

PSNR gains of 0.12-0.14 dB across benchmarks

Up to 29% improvement in frequency-domain consistency

Effective in reducing aliasing and preserving fine details

Abstract

We propose Frequency-Guided Attention (FGA), a lightweight upsampling module for single image super-resolution. Conventional upsamplers, such as Sub-Pixel Convolution, are efficient but frequently fail to reconstruct high-frequency details and introduce aliasing artifacts. FGA addresses these issues by integrating (1) a Fourier feature-based Multi-Layer Perceptron (MLP) for positional frequency encoding, (2) a cross-resolution Correlation Attention Layer for adaptive spatial alignment, and (3) a frequency-domain L1 loss for spectral fidelity supervision. Adding merely 0.3M parameters, FGA consistently enhances performance across five diverse super-resolution backbones in both lightweight and full-capacity scenarios. Experimental results demonstrate average PSNR gains of 0.12~0.14 dB and improved frequency-domain consistency by up to 29%, particularly evident on texture-rich datasets. Visual and spectral evaluations confirm FGA's effectiveness in reducing aliasing and preserving fine details, establishing it as a practical, scalable alternative to traditional upsampling methods.