Research map / Detection and segmentation / CNN features
CBAM: Convolutional Block Attention Module
Sanghyun Woo, Jongchan Park, Joon-Young Lee, In So Kweon
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add max-pooled features to average-pooled channel attention (ablation) | ResNet50 + channel attention · ImageNet-1K val | Top-1 error (%): 23.14 -> 22.80 | Improved |
| add max-pooled features to average-pooled channel attention (ablation) | ResNet50 + channel attention · ImageNet-1K val | Top-5 error (%): 6.70 -> 6.52 | Improved |
| add spatial attention after channel attention (ablation) | ResNet50 + CBAM · ImageNet-1K val | Top-1 error (%): 22.80 -> 22.66 | Improved |
| compute spatial attention with channel avg and max pooling instead of 1x1 conv (ablation) | ResNet50 + CBAM (k=7) · ImageNet-1K val | Top-1 error (%): 22.90 -> 22.66 | Improved |
| use 7x7 instead of 3x3 kernel in spatial attention (ablation) | ResNet50 + CBAM (avg&max) · ImageNet-1K val | Top-1 error (%): 22.68 -> 22.66 | Improved |
| arrange channel and spatial attention sequentially instead of in parallel (ablation) | ResNet50 + CBAM · ImageNet-1K val | Top-1 error (%): 22.95 -> 22.66 | Improved |
| put channel attention first instead of spatial first (ablation) | ResNet50 + CBAM · ImageNet-1K val | Top-1 error (%): 22.78 -> 22.66 | Improved |
| add CBAM to ResNet50 | ResNet50 · ImageNet-1K val | Top-1 error (%): 24.56 -> 22.66 | Improved |
| add CBAM instead of SE to ResNet101 | ResNet101 + SE · ImageNet-1K val | Top-1 error (%): 22.35 -> 21.51 | Improved |
| add CBAM instead of SE to ResNeXt50 | ResNeXt50 + SE · ImageNet-1K val | Top-1 error (%): 21.91 -> 21.92 | No clear effect |
| add CBAM to MobileNet | MobileNet · ImageNet-1K val | Top-1 error (%): 31.39 -> 29.01 | Improved |
| add CBAM to Faster R-CNN backbone | Faster R-CNN ResNet101 · MS COCO val | mAP@[.5,.95]: 29.1 -> 30.8 | Improved |
Closest papers by meaning
- Learn to Pay Attention (ICLR 2018 · similarity 0.90)
- SimAM: A Simple, Parameter-Free Attention Module for Convolutional Neural Networks (ICML 2021 · similarity 0.89)
- TDAM: Top-Down Attention Module for Contextually Guided Feature Selection in CNNs (ECCV 2022 · similarity 0.88)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.86)
- Faster Neighborhood Attention: Reducing the O(n^2) Cost of Self Attention at the Threadblock Level (NeurIPS 2024 · similarity 0.85)
- Bridging the Divide: Reconsidering Softmax and Linear Attention (NeurIPS 2024 · similarity 0.85)
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.
