{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/emulating-self-attention-with-convolution-for","title":"Emulating Self-attention with Convolution for Efficient Image Super-Resolution","arxiv_id":"2503.06671","date":"2025-03-09","proceeding":null,"authors":["Dongheon Lee","Seokju Yun","Youngmin Ro"],"abstract":"In this paper, we tackle the high computational overhead of transformers for lightweight image super-resolution. (SR). Motivated by the observations of self-attention's inter-layer repetition, we introduce a convolutionized self-attention module named Convolutional Attention (ConvAttn) that emulates self-attention's long-range modeling capability and instance-dependent weighting with a single shared large kernel and dynamic kernels. By utilizing the ConvAttn module, we significantly reduce the reliance on self-attention and its involved memory-bound operations while maintaining the representational capability of transformers. Furthermore, we overcome the challenge of integrating flash attention into the lightweight SR regime, effectively mitigating self-attention's inherent memory bottleneck. We scale up window size to 32$\\times$32 with flash attention rather than proposing an intricated self-attention module, significantly improving PSNR by 0.31dB on Urban100$\\times$2 while reducing latency and memory usage by 16$\\times$ and 12.2$\\times$. Building on these approaches, our proposed network, termed Emulating Self-attention with Convolution (ESC), notably improves PSNR by 0.27 dB on Urban100$\\times$4 compared to HiT-SRF, reducing the latency and memory usage by 3.7$\\times$ and 6.2$\\times$, respectively. Extensive experiments demonstrate that our ESC maintains the ability for long-range modeling, data scalability, and the representational power of transformers despite most self-attentions being replaced by the ConvAttn module.","url_abs":"https://arxiv.org/abs/2503.06671v1","url_pdf":"https://arxiv.org/pdf/2503.06671v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"emulating-self-attention-with-convolution-for","repo_url":"https://github.com/dslisleedh/ESC","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"computational-efficiency","task_name":"Computational Efficiency"},{"task_slug":"image-super-resolution","task_name":"Image Super-Resolution"},{"task_slug":"long-range-modeling","task_name":"Long-range modeling"},{"task_slug":"super-resolution","task_name":"Super-Resolution"}],"methods":[{"method_slug":"attention","method_name":"Attention"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2503.06671","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}