Research map / Image generation / Super-resolution
SRFormer: Permuted Self-Attention for Single Image Super-Resolution
Yupeng Zhou, Zhen Li, Chun-Le Guo, Song Bai, Ming-Ming Cheng, Qibin Hou
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 |
|---|---|---|---|
| enlarge window size from 12x12 to 24x24 (ablation) | SRFormer · Urban100 (x2) | PSNR (dB): 33.08 -> 33.51 | Improved |
| enlarge window size from 8x8 to 16x16 (ablation) | SwinIR · Urban100 (x2) | PSNR (dB): 33.09 -> 33.40 | Improved |
| enlarge window from 12x12 to 24x24 without ConvFFN (ablation) | SRFormer w/o ConvFFN · Urban100 (x2) | PSNR (dB): 32.99 -> 33.38 | Improved |
| add ConvFFN (5x5 depth-wise conv in FFN) (ablation) | SRFormer 24x24 · Urban100 (x2) | PSNR (dB): 33.38 -> 33.51 | Improved |
| add 5x5 depth-wise conv in ConvFFN (ablation) | SRFormer · Manga109 (x2) | PSNR (dB): 39.44 -> 39.49 | Improved |
| use 3x3 instead of no depth-wise conv (ablation) | SRFormer · Manga109 (x2) | PSNR (dB): 39.44 -> 39.34 | Hurt |
| use 5x5 instead of 3x3 depth-wise conv (ablation) | SRFormer · Urban100 (x2) | PSNR (dB): 33.42 -> 33.51 | Improved |
| replace token reduction with permuted self-attention (ablation) | SRFormer (S=24) · Urban100 (x2) | PSNR (dB): 33.24 -> 33.38 | Improved |
| replace random token sampling with permuted self-attention (ablation) | SRFormer (S=24) · Urban100 (x2) | PSNR (dB): 32.34 -> 33.38 | Improved |
| use PSA with 24x24 window instead of SwinIR 8x8 window | SRFormer · Urban100 (x2) | parameters and MACs: 11.75M and 2868G -> 10.06M and 2703G | Improved |
| replace SwinIR with SRFormer (DIV2K-trained) | SRFormer · Urban100 (x2) | PSNR (dB): 33.40 -> 33.86 | Improved |
| add self-ensemble (SRFormer+) | SRFormer · Urban100 (x2, DF2K) | PSNR (dB): 34.09 -> 34.29 | Improved |
Later papers that compared with this method
4 later accepted papers report SRFormer as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- MP-HSIR: A Multi-Prompt Framework for Universal Hyperspectral Image Restoration (ICCV 2025 · better on 1 of 1 shared result)
- MambaIR: A Simple Baseline for Image Restoration with State-Space Model (ECCV 2024 · better on 2 of 3 shared results)
- Efficient Cascaded Multiscale Adaptive Network for Image Restoration (ECCV 2024 · better on 1 of 2 shared results)
- A New Dataset and Framework for Real-World Blurred Images Super-Resolution (ECCV 2024 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · similarity 0.89)
- Transcending the Limit of Local Window: Advanced Super-Resolution Transformer with Adaptive Token Dictionary (CVPR 2024 · similarity 0.88)
- Efficient Attention-Sharing Information Distillation Transformer for Lightweight Single Image Super-Resolution (AAAI 2025 · similarity 0.87)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 2024 · similarity 0.87)
- CFAT: Unleashing Triangular Windows for Image Super-resolution (CVPR 2024 · similarity 0.86)
- Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · similarity 0.86)
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.
