AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Scalable Vision Transformers With Hierarchical Pooling

Zizheng Pan, Bohan Zhuang, Jing Liu, Haoyu He, Jianfei Cai

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 175 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
add hierarchical pooling (1 stage) to DeiT-TiDeiT-Ti · ImageNetTop-1 accuracy: 72.20 -> 69.64Hurt
add hierarchical pooling (1 stage) to DeiT-SDeiT-S · ImageNetTop-1 accuracy: 79.80 -> 78.00Hurt
scale up embedding dim and heads with saved FLOPs, 4 pooling stagesHVT-Ti · ImageNetTop-1 accuracy: 72.20 -> 75.23Improved
remove class token and predict from average-pooled tokens (ablation)DeiT-Ti · CIFAR-100Top-1 accuracy: 64.49 -> 65.93Improved
remove class token and predict from average-pooled tokens (ablation)DeiT-Ti · ImageNetTop-1 accuracy: 72.20 -> 72.42Improved
use max pooling instead of average pooling (ablation)HVT-S-4 · CIFAR-100Top-1 accuracy: 70.38 -> 75.43Improved
use max pooling instead of convolution downsampling (ablation)HVT-S-4 · CIFAR-100Top-1 accuracy: 69.75 -> 75.43Improved
increase pooling stages from 0 to 4 (ablation)HVT-S · CIFAR-100Top-1 accuracy: 71.99 -> 75.43Improved
increase pooling stages from 0 to 4 (ablation)HVT-S · ImageNetTop-1 accuracy: 80.39 -> 75.23Hurt
increase Transformer blocks from 12 to 24 (ablation)HVT-S-4 · CIFAR-100Top-1 accuracy: 75.43 -> 75.04No clear effect
increase attention heads from 3 to 16 (ablation)HVT-Ti-4 · CIFAR-100Top-1 accuracy: 69.51 -> 76.30Improved
increase image resolution from 160 to 384 (ablation)HVT-S-4 · CIFAR-100Top-1 accuracy: 73.84 -> 76.31Improved

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