AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Vision Transformer With Progressive Sampling

Xiaoyu Yue, Shuyang Sun, Zhanghui Kuang, Meng Wei, Philip H.S. Torr, Wayne Zhang, Dahua Lin

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 133 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
increase sampling points per axis from n=10 to n=18 (ablation)PS-ViT-B · ImageNet-1ktop-1 accuracy (%): 80.6 -> 82.3Improved
increase iterations of progressive sampling from N=1 (fixed locations) to N=4 (ablation)PS-ViT-B/14 · ImageNet-1ktop-1 accuracy (%): 80.6 -> 81.7Improved
increase iterations from N=4 to N=8 (ablation)PS-ViT-B/14 · ImageNet-1ktop-1 accuracy (%): 81.7 -> 81.9No clear effect
increase iterations beyond N=8 (fewer transformer layers remain) (ablation)PS-ViT-B/14 · ImageNet-1ktop-1 accuracy (%): 81.9 -> 81.6Hurt
add progressive sampling module in place of ViT hard patch splitting (ablation)ViT with same configuration · ImageNet-1ktop-1 accuracy (%): 78.4 -> 81.7Improved
share weights across progressive sampling iterations (ablation)PS-ViT-B/14 · ImageNet-1ktop-1 accuracy (%): 81.7 -> 81.5Hurt
share weights across progressive sampling iterations (ablation)PS-ViT-Ti/14 · ImageNet-1ktop-1 accuracy (%): 75.6 -> 74.1Hurt
share weights across progressive sampling iterations (ablation)PS-ViT-B/10 · ImageNet-1ktop-1 accuracy (%): 80.6 -> 80.0Hurt
use PS-ViT-B/14 instead of DeiT-B for transferPS-ViT-B/14 · Flowers-102top-1 accuracy (%): 98.4 -> 98.8Improved
use PS-ViT-B/14 instead of DeiT-B for transferPS-ViT-B/14 · Stanford Carstop-1 accuracy (%): 92.1 -> 92.9Improved
use PS-ViT-B/14 instead of DeiT-B for transferPS-ViT-B/14 · CIFAR-100top-1 accuracy (%): 90.8 -> 90.8No clear effect
use PS-ViT-B/14 instead of DeiT-B at similar accuracyPS-ViT-B/14 · Speed (V100)images per second: 316.1 -> 765.6Improved

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