Scholarlink

<p>Transformer-based methods have made significant progress in the field of image super-resolution reconstruction. While these methods perform well in capturing the global background, they sometimes struggle to retain detailed information in the reconstructed image. To address the above problem, we propose a mixed multiscale residual attention network (MMRANet). Specifically, we propose hybrid attention (MAB) and multiscale convolutionally gated linear unit (MCGLU). They enable the network to efficiently capture spatial structural information and features at different scales of an image. In addition, we propose a mixed upsampling that further improves the detail recovery of the image. Finally, we introduce overlapping cross-attention to help us better exchange window information. Experimental results on five benchmark datasets show that the method can achieve superior performance and outperform many state-of-the-art reconstruction methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mixed multi-scale residual attention networks for single image super-resolution reconstruction

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • Liyun Zhang,
  • Ming Zhang,
  • Fei Fan,
  • Yang Liu

摘要

Transformer-based methods have made significant progress in the field of image super-resolution reconstruction. While these methods perform well in capturing the global background, they sometimes struggle to retain detailed information in the reconstructed image. To address the above problem, we propose a mixed multiscale residual attention network (MMRANet). Specifically, we propose hybrid attention (MAB) and multiscale convolutionally gated linear unit (MCGLU). They enable the network to efficiently capture spatial structural information and features at different scales of an image. In addition, we propose a mixed upsampling that further improves the detail recovery of the image. Finally, we introduce overlapping cross-attention to help us better exchange window information. Experimental results on five benchmark datasets show that the method can achieve superior performance and outperform many state-of-the-art reconstruction methods.