Research map / Architectures and efficiency / Attention and transformers
Fast and Accurate Image Restoration with Rank Enhanced Linear Attention
Yuang Ai
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 rank-enhancing depthwise conv (RELA) to vanilla linear attention (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 39.28 -> 41.17 (+1.89 dB) | Improved |
| use RELA instead of window self-attention (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.76 -> 41.17 | Improved |
| use RELA instead of transposed self-attention (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.52 -> 41.17 | Improved |
| add channel attention block (CAB) (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.81 -> 41.17 | Improved |
| use CG-FFN instead of MLP (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.20 -> 41.17 | Improved |
| add depthwise conv in CG-FFN (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.11 -> 41.17 | Improved |
| add GLU gating in CG-FFN (ablation) | LAformer-T · SOTS-Indoor dehazing | PSNR (dB): 40.86 -> 41.17 | Improved |
| use 1+ELU instead of ReLU as activation in RELA (ablation) | LAformer-B · GoPro motion deblurring | PSNR (dB): 33.38 -> 33.40 | No clear effect |
| enlarge DWC kernel from 3x3 to 5x5 (ablation) | LAformer-B · GoPro motion deblurring | PSNR (dB): 33.35 -> 33.40 | Improved |
| enlarge DWC kernel from 5x5 to 7x7 (ablation) | LAformer-B · GoPro motion deblurring | PSNR (dB): 33.40 -> 33.41 | No clear effect |
| use LAformer instead of DiT-S/2 for diffusion generation | DiT-S/2 · ImageNet 256 class-conditional diffusion (400K steps) | FID: 68.40 -> 50.39 | Improved |
| use LAformer instead of DiT-XL/2 for diffusion generation | DiT-XL/2 · ImageNet 256 class-conditional diffusion (400K steps) | FID: 19.47 -> 11.31 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Stripformer — Stripformer: Strip Transformer for Fast Image Deblurring (ECCV 2022) · better on 1 of 1 shared result
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
- Breaking the Low-Rank Dilemma of Linear Attention (CVPR 2025 · similarity 0.91)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.90)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.90)
- ACL: Activating Capability of Linear Attention for Image Restoration (CVPR 2025 · similarity 0.89)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 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.
