Research map / Architectures and efficiency / Attention and transformers
Augmented Shortcuts for Vision Transformers
Yehui Tang, Kai Han, Chang Xu, An Xiao, Yiping Deng, Chao Xu, Yunhe Wang
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 augmented shortcuts to DeiT-S | ViT(DeiT)-S · ImageNet val | top-1 accuracy: 79.8 -> 80.9 | Improved |
| Add augmented shortcuts to DeiT-B at 384 resolution | ViT(DeiT)-B · ImageNet val | top-1 accuracy: 83.1 -> 84.2 | Improved |
| Add augmented shortcuts to PVT-M | PVT-M · ImageNet val | top-1 accuracy: 81.2 -> 82.3 | Improved |
| Add augmented shortcuts to T2T-14 | T2T-14 · ImageNet val | top-1 accuracy: 82.3 -> 83.6 | Improved |
| Increase depth T2T-14 to T2T-24 without augmented shortcuts | T2T · ImageNet val | top-1 accuracy: 81.5 -> 82.3 | Improved |
| Increase depth T2T-14 to T2T-24 with augmented shortcuts | Aug-T2T · ImageNet val | top-1 accuracy: 82.1 -> 83.6 | Improved |
| Increase number of augmented shortcuts from 1 to 2 (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 80.6 -> 80.9 | Improved |
| Increase number of augmented shortcuts from 2 to 3 (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 80.9 -> 80.9 | No clear effect |
| Place augmented shortcuts on both MSA and MLP instead of MSA only (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 80.6 -> 80.9 | Improved |
| Place augmented shortcuts on MLP only (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 79.8 -> 80.4 | Improved |
| Increase block-circulant partitions b from 1 to 4 (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 80.5 -> 80.9 | Improved |
| Increase b from 4 to 8 (ablation) | Aug-ViT-S · ImageNet val | top-1 accuracy: 80.9 -> 80.9 | No clear effect |
Closest papers by meaning
- Visual Transformers: Where Do Transformers Really Belong in Vision Models? (ICCV 2021 · similarity 0.86)
- AdaViT: Adaptive Vision Transformers for Efficient Image Recognition (CVPR 2022 · similarity 0.86)
- The Principle of Diversity: Training Stronger Vision Transformers Calls for Reducing All Levels of Redundancy (CVPR 2022 · similarity 0.86)
- A Survey on Vision Transformer (TPAMI 2022 · similarity 0.86)
- Interpreting vision transformers via residual replacement model (NeurIPS 2025 · similarity 0.85)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · 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.
