AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Local Relation Networks for Image Recognition

Han Hu, Zheng Zhang, Zhenda Xie, Stephen Lin

ICCV 2019 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 639 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
restrict aggregation to a local 7x7 neighborhood instead of full image (ablation)LR-Net-26 (no geometric prior) · ImageNet valtop-1 accuracy: 50.7 -> 71.9Improved
restrict aggregation to a local 7x7 neighborhood instead of full image (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 68.4 -> 75.7Improved
add geometric prior term (small network on relative position) (ablation)LR-Net-26 7x7 · ImageNet valtop-1 accuracy: 71.9 -> 75.7Improved
compute geometric prior with small network instead of directly learned values (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 72.3 -> 75.4 at 9x9Improved
increase kernel size from 3x3 to 7x7 (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 73.6 -> 74.9 -> 75.7Improved
increase convolution kernel size from 5x5 to 9x9 (ablation)ResNet-26 · ImageNet valtop-1 accuracy: 73.0 -> 71.4Hurt
use scalar query/key instead of 16-dim vectors (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 73.7 -> 75.7Improved
share channels (m=8) in aggregation weights (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 75.3 -> 75.7No clear effect
add softmax normalization over spatial scope (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 74.8 -> 75.7Improved
change appearance composability function (squared diff, abs diff, product) (ablation)LR-Net-26 · ImageNet valtop-1 accuracy: 75.7, 75.5, 75.7No clear effect
replace 3x3 convolution with non-local moduleResNet-26 · ImageNet valtop-1 accuracy: 72.8 -> 47.7Hurt
add non-local modules after every blockResNet-26 · ImageNet valtop-1 accuracy: 72.8 -> 73.4Improved

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