Research map / Architectures and efficiency / Attention and transformers
Uformer: A General U-Shaped Transformer for Image Restoration
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace convolutional ResBlocks with LeWin Transformer blocks (ablation) | Uformer on SIDD · SIDD | PSNR: 39.62 -> 39.66 | Helped |
| replace convolutional ResBlocks with LeWin Transformer blocks (ablation) | Uformer-B on SIDD · SIDD | PSNR: 39.71 -> 39.89 | Helped |
| add locality enhancement to the FFN (ablation) | Uformer-S · SIDD | PSNR: 39.74 -> 39.77 | Helped |
| add locality enhancement to W-MSA (ablation) | Uformer-S · SIDD | PSNR: 39.74 -> 39.72 | Hurt |
| add locality enhancement to FFN (ablation) | Uformer-B · RealBlur-R | PSNR: 36.15 -> 36.22 | Helped |
| add locality enhancement to W-MSA as well as FFN (ablation) | Uformer-B · RealBlur-R | PSNR: 36.22 -> 36.19 | Hurt |
| add multi-scale restoration modulator (ablation) | Uformer-T · GoPro | PSNR: 29.11 -> 29.57 | Helped |
| add multi-scale restoration modulator (ablation) | Uformer-B · SIDD | PSNR: 39.86 -> 39.89 | Helped |
| add multi-scale restoration modulator (ablation) | Uformer-B · SPAD | PSNR: 47.43 -> 47.84 | Helped |
| add shifted windows (ablation) | Uformer-S · SIDD | PSNR: 39.76 -> 39.77 | Helped |
| replace feature-concatenation skips with cross-attention skips (ablation) | Uformer-S · SIDD | PSNR: 39.77 -> 39.75 | Hurt |
| replace feature-concatenation skips with concatenated cross-attention skips (ablation) | Uformer-S · SIDD | PSNR: 39.77 -> 39.73 | Hurt |
Closest papers by meaning
- Correlation Matching Transformation Transformers for UHD Image Restoration (AAAI 2024 · similarity 0.87)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.85)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.84)
- Xformer: Hybrid X-Shaped Transformer for Image Denoising (ICLR 2024 · similarity 0.84)
- DreamUHD: Frequency Enhanced Variational Autoencoder for Ultra-High-Definition Image Restoration (AAAI 2025 · similarity 0.84)
- InstructRestore: Region-Customized Image Restoration with Human Instructions (NeurIPS 2025 · similarity 0.84)
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.