Research map / Architectures and efficiency / Attention and transformers
MaxViT: Multi-axis Vision Transformer
Zhengzhong Tu, Hossein Talebi, Han Zhang, Feng Yang, Peyman Milanfar, Alan Bovik, Yinxiao Li
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 grid-attention in stage 1 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.36 -> 83.62 | Improved |
| add grid-attention in stage 2 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.38 -> 83.62 | Improved |
| add grid-attention in stage 3 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.00 -> 83.62 | Improved |
| use grid-attention rather than block-attention in stage 1 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.49 -> 83.62 | Improved |
| use grid-attention rather than block-attention in stage 3 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.40 -> 83.62 | Improved |
| add MBConv in stage 1 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.24 -> 83.62 | Improved |
| add MBConv in stage 3 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 82.65 -> 83.62 | Improved |
| add MBConv in stage 4 (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.09 -> 83.62 | Improved |
| place MBConv before block- and grid-attention (C-BA-GA) (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.07 -> 83.62 | Improved |
| order block-attention before grid-attention (C-BA-GA) instead of C-GA-BA (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 83.54 -> 83.62 | No clear effect |
| use sequential instead of parallel multi-axis attention (ablation) | MaxViT-T · ImageNet-1K | top-1 accuracy: 82.64 -> 83.62 | Improved |
| use sequential instead of parallel multi-axis attention (ablation) | MaxViT-L · ImageNet-1K | top-1 accuracy: 83.54 -> 85.17 | 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.
- CoAtNet-2 — CoAtNet: Marrying Convolution and Attention for All Data Sizes (NeurIPS 2021) · better on 4 of 5 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
- RegionViT: Regional-to-Local Attention for Vision Transformers (ICLR 2022 · similarity 0.92)
- Focal Attention for Long-Range Interactions in Vision Transformers (NeurIPS 2021 · similarity 0.92)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.91)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · similarity 0.90)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.90)
- Breaking the Low-Rank Dilemma of Linear Attention (CVPR 2025 · similarity 0.89)
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.
