Research map / Architectures and efficiency / Attention and transformers
DSwinIR: Rethinking Window-Based Attention for Image Restoration
Gang Wu, Junjun Jiang, Kui Jiang, Xianming Liu, Liqiang Nie
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 token-centric sliding window attention (ablation) | window-attention baseline · 3-task all-in-one avg | PSNR (dB): 30.66 -> 31.79 | Improved |
| add content-aware deformable window (ablation) | window-attention baseline · 3-task all-in-one avg | PSNR (dB): 31.26 -> 32.11 | Improved |
| combine sliding and deformable (DSwin, K=7) (ablation) | DSwinIR · 3-task all-in-one avg | PSNR (dB): 32.11 -> 32.53 | Improved |
| increase kernel size from 5 to 7 (ablation) | DSwin · 3-task all-in-one avg | PSNR (dB): 32.02 -> 32.53 | Improved |
| increase kernel size from 7 to 9 (ablation) | DSwin · 3-task all-in-one avg | PSNR (dB): 32.53 -> 32.61 | No clear effect |
| add multi-scale kernels in one attention module (ablation) | DSwin · 3-task all-in-one avg | PSNR (dB): 32.61 -> 32.69 | Improved |
| add multi-scale gated FFN (ablation) | DSwinIR · 3-task all-in-one avg | PSNR (dB): 32.69 -> 32.72 | Improved |
| use DSwinIR instead of Gridformer backbone | all-in-one · 3-task benchmark | PSNR (dB): 32.19 -> 32.72 | Improved |
| use DSwinIR instead of AdaIR | all-in-one · 5-task benchmark | PSNR (dB): 30.20 vs 30.20 | No clear effect |
| use DSwinIR instead of PromptIR | all-in-one · 3-task benchmark latency | PSNR (dB): 0.71s -> 0.55s | Improved |
Closest papers by meaning
- Fast and Accurate Image Restoration with Rank Enhanced Linear Attention (ECCV 2026 · similarity 0.89)
- HAT: Hybrid Attention Transformer for Image Restoration (TPAMI 2025 · similarity 0.89)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.86)
- Intra and Inter Parser-Prompted Transformers for Effective Image Restoration (AAAI 2025 · similarity 0.86)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.85)
- ReFIR: Grounding Large Restoration Models with Retrieval Augmentation (NeurIPS 2024 · similarity 0.85)
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.
