Research map / Architectures and efficiency / Attention and transformers
Adaptive Token Sampling for Efficient Vision Transformers
Mohsen Fayyaz, Soroush Abbasi Koohpayegani, Farnoush Rezaei Jafari, Sunando Sengupta, Hamid Reza Vaezi Joze, Eric Sommerlade, Hamed Pirsiavash, Jürgen Gall
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 ATS to DeiT-S (stages 3-11) with fine-tuning | DeiT-S · ImageNet-1K val | GFLOPs: 4.6 -> 2.9 | Improved |
| add ATS to DeiT-S (stages 3-11) with fine-tuning | DeiT-S · ImageNet-1K val | top-1 accuracy (%): 79.8 -> 79.7 | No clear effect |
| add ATS to CvT-21 | CvT-21 · ImageNet-1K val | GFLOPs: 7.1 -> 5.1 | Improved |
| add ATS to TimeSformer-L | TimeSformer-L · Kinetics-400 | GFLOPs: 7,140 -> 3,510 | Improved |
| add ATS to X-ViT 16x | X-ViT · Kinetics-400 | GFLOPs: 425 -> 259 | Improved |
| add ATS to EViT-DeiT-S | EViT-DeiT-S · ImageNet-1K val | GFLOPs: 3.0 -> 2.5 | Improved |
| use classification-token attention rather than summed self-attention as significance score (ablation) | ATS DeiT-S · ImageNet-1K val | top-1 accuracy (%) | Improved |
| use classification-token attention rather than a random token's attention (ablation) | ATS DeiT-S · ImageNet-1K val | top-1 accuracy (%) | Improved |
| add L2 norm of values to the significance score (ablation) | ATS DeiT-S · ImageNet-1K val | top-1 accuracy (%): about 0.2% | Improved |
| use inverse transform sampling instead of Top-K selection (ablation) | ATS DeiT-S · ImageNet-1K val | top-1 accuracy (%) | Improved |
| fine-tune DeiT-S+ATS (ablation) | DeiT-S+ATS · ImageNet-1K val | top-1 accuracy (%) | Improved |
| fine-tune XViT+ATS (ablation) | XViT+ATS 16x · Kinetics-600 | top-1 accuracy (%): 83.4 -> 84.4 | Improved |
Closest papers by meaning
- A-ViT: Adaptive Tokens for Efficient Vision Transformer (CVPR 2022 · similarity 0.88)
- Not All Images are Worth 16x16 Words: Dynamic Transformers for Efficient Image Recognition (NeurIPS 2021 · similarity 0.88)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 2024 · similarity 0.86)
- A General and Efficient Training for Transformer via Token Expansion (CVPR 2024 · similarity 0.86)
- Multi-criteria Token Fusion with One-step-ahead Attention for Efficient Vision Transformers (CVPR 2024 · similarity 0.86)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.86)
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.
