AI Paper Atlas

Research map / Detection and segmentation / CNN features

CBAM: Convolutional Block Attention Module

Sanghyun Woo, Jongchan Park, Joon-Young Lee, In So Kweon

ECCV 2018 · Detection and segmentation › CNN features · approach: Efficient network training and design · 26,917 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 max-pooled features to average-pooled channel attention (ablation)ResNet50 + channel attention · ImageNet-1K valTop-1 error (%): 23.14 -> 22.80Improved
add max-pooled features to average-pooled channel attention (ablation)ResNet50 + channel attention · ImageNet-1K valTop-5 error (%): 6.70 -> 6.52Improved
add spatial attention after channel attention (ablation)ResNet50 + CBAM · ImageNet-1K valTop-1 error (%): 22.80 -> 22.66Improved
compute spatial attention with channel avg and max pooling instead of 1x1 conv (ablation)ResNet50 + CBAM (k=7) · ImageNet-1K valTop-1 error (%): 22.90 -> 22.66Improved
use 7x7 instead of 3x3 kernel in spatial attention (ablation)ResNet50 + CBAM (avg&max) · ImageNet-1K valTop-1 error (%): 22.68 -> 22.66Improved
arrange channel and spatial attention sequentially instead of in parallel (ablation)ResNet50 + CBAM · ImageNet-1K valTop-1 error (%): 22.95 -> 22.66Improved
put channel attention first instead of spatial first (ablation)ResNet50 + CBAM · ImageNet-1K valTop-1 error (%): 22.78 -> 22.66Improved
add CBAM to ResNet50ResNet50 · ImageNet-1K valTop-1 error (%): 24.56 -> 22.66Improved
add CBAM instead of SE to ResNet101ResNet101 + SE · ImageNet-1K valTop-1 error (%): 22.35 -> 21.51Improved
add CBAM instead of SE to ResNeXt50ResNeXt50 + SE · ImageNet-1K valTop-1 error (%): 21.91 -> 21.92No clear effect
add CBAM to MobileNetMobileNet · ImageNet-1K valTop-1 error (%): 31.39 -> 29.01Improved
add CBAM to Faster R-CNN backboneFaster R-CNN ResNet101 · MS COCO valmAP@[.5,.95]: 29.1 -> 30.8Improved

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