Lightweight Single Image Super-Resolution With High-Continuity Attention

• Published in: IEEE signal processing letters Vol. 32; pp. 2614 - 2618
• Main Authors: Zhang, Ju, Zhong, Baojiang, Ma, Kai-Kuang
• Format: Journal Article
• Language: English
• Published: IEEE 2025
• Subjects: and high continuity; Computational efficiency; Computational modeling; Computer architecture; Convolution; Data mining; Feature extraction; Image reconstruction; Iron; Scale-dual attention; single image super-resolution (SISR); Superresolution; Training; window attention
• ISSN: 1070-9908; 1558-2361
• DOI: 10.1109/LSP.2025.3584016

Window attention has become a popular choice in single image super-resolution (SISR) network design due to its efficient computation. However, its self-attention is restricted to fixed-size windows, leading to a lack of cross-window interaction. To address this, the benchmark SwinIR model adopts a shifted window strategy to capture long-range dependencies. However, we observe that its attention still suffers from discontinuities at window boundaries, resulting in inferior SISR performance. To address this issue, we propose a new scale-dual attention (SDA) module, consisting of three parallel branches that integrate window attention and pooling attention via three complementary scales. This enables hierarchical local-global interactions, yielding high-continuity attention maps. To validate the effectiveness of our proposed SDA, we develop a lightweight scale-dual attention network (SDAN) with approximately 878 K parameters for SISR. Extensive experiments demonstrate that our SDAN achieves superior performance, outperforming state-of-the-art methods in both accuracy and efficiency.