Research map / Image generation / Image restoration
MB-TaylorFormer V2: Improved Multi-Branch Linear Transformer Expanded by Taylor Formula for Image Restoration
Zhi Juan Jin, Yuwei Qiu, Kaihao Zhang, Hongdong Li, Wenhan Luo
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 convolutional positional encoding (CPE) (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 39.76 -> 41.00 | Improved |
| use CPE instead of MSAR (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 40.73 -> 41.00 | Improved |
| use T-MSA++ instead of MDTA channel attention (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 38.57 -> 41.00 | Improved |
| use T-MSA++ instead of Swin window attention (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 36.59 -> 41.00 | Improved |
| use T-MSA++ instead of Flatten attention (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 40.47 -> 41.00 | Improved |
| add parallel branches with single-scale conv patch embedding (Conv-P) (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 39.43 -> 39.87 | Improved |
| replace standard conv with DSDCN in serial-parallel patch embedding (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 40.25 -> 41.00 | Improved |
| truncate DSDCN offsets to [-3, 3] (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 40.13 -> 41.00 | Improved |
| widen offset truncation range to [-4, 4] (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 41.00 -> 40.53 | Hurt |
| vary focused factor p from 3 to 8 (ablation) | MB-TaylorFormer-B V2 · ITS dehazing | PSNR: 40.81, 41.00, 40.93, 40.76 | No clear effect |
| add second-order Taylor remainder approximation to Swin (ablation) | Swin attention · ITS dehazing | PSNR: 36.59 -> 36.50 | No clear effect |
| upgrade V1 to V2 (T-MSA++ and CPE) | MB-TaylorFormer-L · deraining (average) | PSNR: 33.66 -> 34.00 | 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.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · 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
- MB-TaylorFormer: Multi-Branch Efficient Transformer Expanded by Taylor Formula for Image Dehazing (ICCV 2023 · similarity 0.93)
- Unfolding Taylor's Approximations for Image Restoration (NeurIPS 2021 · similarity 0.89)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.87)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 2024 · similarity 0.87)
- Correlation Matching Transformation Transformers for UHD Image Restoration (AAAI 2024 · similarity 0.87)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (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.
