AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Group Normalization

Yuxin Wu, Kaiming He

ECCV 2018 · Architectures and efficiency › RNNs and normalization · approach: Batch normalization improvements · 4,036 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 BN with GN at batch size 2 (ablation)ResNet-50 · ImageNet valtop-1 error (%): 34.7 -> 24.1Improved
replace BN with GN at batch size 32ResNet-50 · ImageNet valtop-1 error (%): 23.6 -> 24.1Hurt
replace BN with GN at batch size 8 (ablation)ResNet-50 · ImageNet valtop-1 error (%): 24.8 -> 24.0Improved
use GN instead of Batch Renorm at batch size 4 (ablation)ResNet-50 · ImageNet valtop-1 error (%): 26.3 -> 24.2Improved
use grouped channels (2 per group) instead of per-channel IN (ablation)GN ResNet-50 · ImageNet valtop-1 error (%): 28.4 -> 25.6Improved
use G=32 groups instead of G=1 (LN) (ablation)GN ResNet-50 · ImageNet valtop-1 error (%): 25.3 -> 24.1Improved
replace BN with GN on ResNet-101 at batch size 2 (ablation)ResNet-101 · ImageNet valtop-1 error (%): 31.9 -> 23.0Improved
replace frozen BN with GN in C4 Mask R-CNN (ablation)Mask R-CNN C4 · COCO val2017box AP: 37.7 -> 38.8Improved
replace GN with LN in C4 Mask R-CNN (ablation)Mask R-CNN C4 · COCO val2017box AP: 1.9 box AP worse than GNHurt
add GN to the box head with BN* backbone (ablation)Mask R-CNN FPN · COCO val2017box AP: 38.6 -> 39.5Improved
add GN to the backbone (pretrained GN model) (ablation)Mask R-CNN FPN · COCO val2017box AP: 39.5 -> 40.0Improved
train GN longer (180k to 270k iterations) (ablation)R50 GN Mask R-CNN FPN · COCO val2017box AP: 40.3 -> 40.8Improved

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