Scholarlink

<p>Low-power edge platforms lack the computational resources required by many state-of-the-art super-resolution (SR) models, limiting their practical deployment. To address this, we introduce SCPFAN, a lightweight and efficient SR network whose core building block is the SCPFARB module, comprising (i) a spatially shifted <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(1\times 1\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>1</mn> <mo>×</mo> <mn>1</mn> </mrow> </math></EquationSource> </InlineEquation> convolution (SSConv) and (ii) parameter-free attention (PFA). SSConv enlarges the receptive field while retaining the computational cost of a standard <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(1\times 1\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>1</mn> <mo>×</mo> <mn>1</mn> </mrow> </math></EquationSource> </InlineEquation> convolution. PFA reweights features on the fly using input statistics, without introducing learnable parameters. Experimental results indicate that the proposed method has a parameter count comparable to current efficient state-of-the-art methods. Code is available at <a href="https://github.com/tu-sinsie/SCPFAN">https://github.com/tu-sinsie/SCPFAN</a>.</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

Efficient image super-resolution via convolutional shift and parameter-free attention mechanisms

  • 期刊论文

  • 发表日期：

    2026年4月26日

查看全文
  • Linlin Wang,
  • Xiaoyan Gao,
  • Yajie Wang,
  • Chuanyun Wang,
  • Qian Gao,
  • Xueyi Xi

摘要

Low-power edge platforms lack the computational resources required by many state-of-the-art super-resolution (SR) models, limiting their practical deployment. To address this, we introduce SCPFAN, a lightweight and efficient SR network whose core building block is the SCPFARB module, comprising (i) a spatially shifted \(1\times 1\) 1 × 1 convolution (SSConv) and (ii) parameter-free attention (PFA). SSConv enlarges the receptive field while retaining the computational cost of a standard \(1\times 1\) 1 × 1 convolution. PFA reweights features on the fly using input statistics, without introducing learnable parameters. Experimental results indicate that the proposed method has a parameter count comparable to current efficient state-of-the-art methods. Code is available at https://github.com/tu-sinsie/SCPFAN.