Progressive feature aggregation network for image super-resolution

Ruilin Zhang

doi:10.1016/j.neucom.2025.132058

Progressive feature aggregation network for image super-resolution

2025-11-07

0

PRE

AI

DOI：10.1016/j.neucom.2025.132058

Original

Original request for help

Save

Abstract

En 中文

Recently, deploying single image super-resolution (SISR) on low-resource devices remains a challenge due to high computational costs, especially for transformer-based self-attention (SA) methods. Their low-pass filtering characteristics limit the ability to capture local features, resulting in overly smooth reconstruction results. To alleviate this problem, in this paper we propose a progressive feature aggregation network (PFANet) for image super-resolution. Concretely, we propose an adaptive feature aggregation module (AFAM). It first uses an efficient approximation method from the wavelet convolution attention (WCA) branch to estimate non-local information, and then the edge-preserving estimation (EPE) branch aggregates local detail features. Finally, progressive skip connection (PSC) enable contextual association between deep information and feature reuse. Comprehensive experiments demonstrate that our PFANet achieves a competitive performance with efficient resource utilization on public benchmarks. In particular, compared to the 4 SwinIR-light, PFANet improves performance by an average of 0.12 dB in five public testsets, while reducing model parameters by approximately 34 %.

Progressive feature aggregation network for image super-resolution

Abstract

Journal

Organization