AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

HorNet: Efficient High-Order Spatial Interactions with Recursive Gated Convolutions

Yongming Rao, Wenliang Zhao, Yansong Tang, Jie Zhou, Ser-Nam Lim, Jiwen Lu

NeurIPS 2022 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 184 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 spatial mixing with recursive gated convolution using global filtersHorNet backbones · ImageNet-1K valtop-1 accuracy: 82.1 -> 83.0Improved
use global filters instead of 7×7 convolutionHorNet-T · ImageNet-1K valtop-1 accuracy: 82.8 -> 83.0Improved
replace 7×7 self-attention with 7×7 DWConv, then add SE (ablation)Swin-T-derived baseline · ImageNet-1K valtop-1 accuracy: 81.2 -> 81.5Improved
replace SE with order-1 recursive gated convolution (ablation)Swin-T-derived baseline · ImageNet-1K valtop-1 accuracy: 81.2 -> 81.7Improved
increase shared interaction order through 3, then 4 (ablation)Swin-T-derived baseline · ImageNet-1K valtop-1 accuracy: 81.7 -> 82.5; 82.5 -> 82.5Mixed
progressively increase interaction order across stages from 2 to 5 (ablation)Swin-T-derived baseline · ImageNet-1K valtop-1 accuracy: 82.5 -> 82.6Improved
replace self-attention with recursive gated convolution (ablation)isotropic DeiT-S · ImageNet-1K valtop-1 accuracy: 79.8 -> 80.6Improved
replace 3×3 depth-wise convolution with recursive gated convolution (ablation)3×3 spatial-mixing model · ImageNet-1K valtop-1 accuracy: 80.7 -> 82.1Improved
replace pooling with recursive gated convolution (ablation)pooling spatial-mixing model · ImageNet-1K valtop-1 accuracy: 78.1 -> 79.3Improved
replace standard FPN fusion convolution with HorFPN global filters (ablation)ResNet-50 dense-prediction models · ADE20K val, UperNet 160Ksingle-scale mIoU: 40.7 -> 43.2Improved
set recursive-gating scale α to 3 rather than 2 (ablation)HorNet-B 7×7 · ImageNet-1K valtop-1 accuracy: 82.76 -> 82.81Improved
add Sigmoid activation to gated-convolution weights (ablation)HorNet gated convolutions · ImageNet-1K valtop-1 accuracy: 82.7 -> 82.3Hurt

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