Lightweight Single Image Super-Resolution With High-Continuity Attention

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 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">scale-dual attention</i> (SDA) module, consisting of three parallel branches that integrate <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">window</i> attention and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">pooling</i> 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 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">scale-dual attention network</i> (SDAN) with approximately 878K parameters for SISR. Extensive experiments demonstrate that our SDAN achieves superior performance, outperforming state-of-the-art methods in both accuracy and efficiency.