Research map / Architectures and efficiency / Attention and transformers
Accurate Image Restoration with Attention Retractable Transformer
Read the paperCheck 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 |
|---|---|---|---|
| alternate dense and sparse attention blocks instead of using only dense or only sparse blocks (ablation) | ART · Urban100 ×2 | PSNR | Helped |
| use only dense attention blocks, removing sparse attention (ablation) | ART · Urban100 ×2 | PSNR | Hurt |
| use only sparse attention blocks, removing dense attention (ablation) | ART · Urban100 ×2 | PSNR | Hurt |
| reduce the sparse-attention interval (ablation) | ART · Urban100 ×2 | PSNR | Helped |
| set MLP ratio to 2 and sparse-attention interval to 8 (ablation) | ART · Urban100 ×2 | PSNR | Helped |
| add self-ensemble at inference | ART · Image SR ×2, ×3, and ×4; denoising; JPEG CAR | PSNR: 33.04 -> 33.07 | Helped |
| use ART instead of SwinIR | Image SR · Urban100 ×4 | PSNR: 27.45 -> 27.77 | Helped |
| use ART instead of Restormer | Image denoising · Kodak24 σ=50 | PSNR: 30.01 -> 29.87 | Hurt |
| use ART instead of SwinIR | JPEG CAR · Classic5 q=10 | PSNR: 30.27 -> 30.27 | No clear effect |
Closest papers by meaning
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.90)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.87)
- Intra and Inter Parser-Prompted Transformers for Effective Image Restoration (AAAI 2025 · similarity 0.87)
- Attention-Only Transformers via Unrolled Subspace Denoising (ICML 2025 · similarity 0.86)
- Progressive Focused Transformer for Single Image Super-Resolution (CVPR 2025 · similarity 0.86)
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 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.