Research map / Image generation / Super-resolution
Emulating Self-attention with Convolution for Efficient Image Super-Resolution
Dongheon Lee, Seokju Yun, Youngmin Ro
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ConvAttn layers to a self-attention-only network (ablation) | ESC · Urban100 x2 | PSNR: 33.23 -> 33.46 | Improved |
| add self-attention (first layer per block) to a ConvAttn-only network (ablation) | ESC · Urban100 x2 | PSNR: 32.91 -> 33.46 | Improved |
| use 13x13 shared large kernel instead of 9x9 (ablation) | ESC · Urban100 x2 | PSNR: 33.42 -> 33.46 | Improved |
| use 13x13 shared large kernel instead of 17x17 (ablation) | ESC · Urban100 x2 | PSNR: 33.40 -> 33.46 | Improved |
| add LK sharing and dynamic kernel (ablation) | ESC · Urban100 x2 | PSNR: 33.37 -> 33.46 | Improved |
| add LK sharing (ablation) | ESC · Urban100 x2 | PSNR: 33.43 -> 33.46 | Improved |
| add dynamic kernel DK (ablation) | ESC · Urban100 x2 | PSNR: 33.36 -> 33.46 | Improved |
| enlarge self-attention window to 32 from 16 with more layers (ablation) | ESC · Urban100 x2 | PSNR: 33.05 -> 33.46 | Improved |
| enlarge self-attention window to 32 from 24 with more layers (ablation) | ESC · Urban100 x2 | PSNR: 33.28 -> 33.46 | Improved |
| use relative position bias with Flash Attention instead of RoPE (ablation) | ESC · Urban100 x2 | PSNR: 32.76 -> 33.46 | Improved |
| add Flash Attention to 32x32 window self-attention | self-attention · 288x288x64 input | latency: 44.7 -> 2.8 | Improved |
| add Flash Attention to 32x32 window self-attention | self-attention · 288x288x64 input | memory (MB): 2717 -> 223 | Improved |
Closest papers by meaning
- Efficient Long-Range Attention Network for Image Super-Resolution (ECCV 2022 · similarity 0.92)
- Single Image Super-Resolution via a Holistic Attention Network (ECCV 2020 · similarity 0.90)
- Efficient Attention-Sharing Information Distillation Transformer for Lightweight Single Image Super-Resolution (AAAI 2025 · similarity 0.90)
- Transcending the Limit of Local Window: Advanced Super-Resolution Transformer with Adaptive Token Dictionary (CVPR 2024 · similarity 0.89)
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · similarity 0.88)
- CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Transformers Up (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
