AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Making Vision Transformers Efficient From a Token Sparsification View

Shuning Chang, Pichao Wang, Ming Lin, Fan Wang, David Junhao Zhang, Rong Jin, Mike Zheng Shou

CVPR 2023 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 41 citations (OpenAlex)

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 changedWhereMeasuredEffect
Replace image tokens with 16 semantic tokensDeiT-S · ImageNet-1K valTop-1 accuracy: 79.8 -> 79.8(+0.0)No clear effect
Replace image tokens with 16 semantic tokensDeiT-S · ImageNet-1K valFLOPs: 4.58 -> 1.91(-58%)Improved
Use 9 semantic tokens per windowSwin-S · ImageNet-1K valTop-1 accuracy: 83.0 -> 83.0(-0.0)No clear effect
Move STGM toward shallower layers (ablation)Swin-T/S/B · ImageNet-1K valTop-1 accuracy: 81.3 -> 81.0(-0.3%); 83.0 -> 82.8(-0.2%); 83.5 -> 83.2(-0.3%)Hurt
Use STViT-R recovery and dumbbell unitsSwin-S backbone · COCO 2017, Cascade Mask R-CNN 3×AP^b_s: 35.2 -> 36.7Improved
Use STViT-R backboneSwin-S · ADE20K valmIoU: 49.3 -> 48.3Hurt
Combine spatial and global initialization (ablation)STViT-DeiT-S · ImageNet-1K valTop-1 accuracy: 79.4 -> 79.8Improved
Replace spatial initialization with global initialization (ablation)STViT-DeiT-S · ImageNet-1K valTop-1 accuracy: 79.8 -> 78.7Hurt
Add learned positional encoding (ablation)STViT-DeiT-S · ImageNet-1K valTop-1 accuracy: 79.8 -> 79.7Hurt
Increase STGM from 2 to 3 transformer layers (ablation)STViT-DeiT-S · ImageNet-1K valTop-1 accuracy: 79.8 -> 79.5Hurt
Add dumbbell units (ablation)STViT-R-Swin-S · COCO 2017AP^b: 51.4 -> 51.8Improved
Add dumbbell units (ablation)STViT-R-Swin-S · COCO 2017AP^m: 44.4 -> 44.7Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy