Research map / Architectures and efficiency / Attention and transformers
Auto-scaling Vision Transformers without Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use As-ViT Large instead of competing backbones | Cascade Mask R-CNN on COCO 2017 · COCO 2017 val | AP: 52.2 -> 52.7 | Helped |
| Use expected length distortion rather than length distortion with curvature | Training-free ViT topology ranking · ImageNet-1k | Kendall-tau correlation: -0.01 -> 0.49 | Helped |
| Use curvature rather than expected length distortion | Training-free ViT topology ranking · ImageNet-1k | Kendall-tau correlation: 0.49 vs -0.49 | Mixed |
| Add the searched seed topology instead of using random topologies (ablation) | Small ViT topology on ImageNet-1k · ImageNet-1k | Top-1 accuracy: 61.4 -> 61.7 | Helped |
| Add complexity-based automated scaling rather than random scaling (ablation) | As-ViT Small · ImageNet-1k | Top-1 accuracy: 80.5 -> 81.2 | Helped |
| Add complexity-based automated scaling rather than random scaling (ablation) | As-ViT Base · ImageNet-1k | Top-1 accuracy: 82.2 -> 82.5 | Helped |
| Add complexity-based automated scaling rather than random scaling (ablation) | As-ViT Large · ImageNet-1k | Top-1 accuracy: 83.2 -> 83.5 | Helped |
| Add progressive 4× and 2× token reduction during early and middle training (ablation) | As-ViT · ImageNet-1k 224×224 | Top1 Acc.: 83.5 -> 82.5 | Hurt |
| Add progressive 4× and 2× token reduction during early and middle training (ablation) | As-ViT · ImageNet-1k 224×224 | FLOPs saving: 100% -> 56.2\% FLOPs saving | Helped |
| Use full-resolution training throughout instead of progressive token reduction (ablation) | As-ViT · ImageNet-1k 224×224 | Top1 Acc.: 82.5 -> 83.5 | Helped |
Closest papers by meaning
- Scaling Vision Transformers (CVPR 2022 · similarity 0.92)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.90)
- Slicing Vision Transformer for Flexible Inference (NeurIPS 2024 · similarity 0.90)
- Vision Transformers with Self-Distilled Registers (NeurIPS 2025 · similarity 0.89)
- A General and Efficient Training for Transformer via Token Expansion (CVPR 2024 · similarity 0.89)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · 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.