Research map / Architectures and efficiency / CNN design
Squeeze-and-Excitation Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SE blocks to ResNet-50 | ResNet-50 · ImageNet 2012 val | top-5 err.: 7.48 -> 6.62 | Helped |
| Add SE blocks to ResNeXt-50 | ResNeXt-50 · ImageNet 2012 val | top-5 err.: 5.90 -> 5.49 | Helped |
| Add SE blocks to MobileNet | MobileNet · ImageNet 2012 val | top-1 err.: 28.4 -> 25.3 | Helped |
| Add SE blocks to ResNet-110 | ResNet-110 · CIFAR-10 test | classification error: 6.37 -> 5.21 | Helped |
| Add SE blocks to ResNet-152 | ResNet-152 · Places365 val | top-5 err.: 11.61 -> 11.01 | Helped |
| Use SE-ResNet-50 as Faster R-CNN trunk | ResNet-50 Faster R-CNN · COCO minival | AP: 38.0 -> 40.4 | Helped |
| Increase reduction ratio from 16 to 32 (ablation) | SE-ResNet-50 · ImageNet | top-1 err.: 22.28 -> 22.72 | Hurt |
| Replace global average pooling with global max pooling (ablation) | SE-ResNet-50 · ImageNet | top-1 err.: 22.28 -> 22.57 | Hurt |
| Replace sigmoid excitation with tanh (ablation) | SE-ResNet-50 · ImageNet | top-1 err.: 22.28 -> 23.00 | Hurt |
| Replace sigmoid excitation with ReLU (ablation) | SE-ResNet-50 · ImageNet | top-1 err.: 22.28 -> 23.47 | Hurt |
| Add SE blocks only at stage 4 (ablation) | ResNet-50 · ImageNet | top-1 err.: 23.30 -> 22.68 | Helped |
| Move SE blocks after residual and identity branch summation (ablation) | SE-ResNet-50 · ImageNet | top-1 err.: 22.28 -> 22.78 | Hurt |
Closest papers by meaning
- Gated Channel Transformation for Visual Recognition (CVPR 2020 · similarity 0.88)
- Linear Context Transform Block (AAAI 2020 · similarity 0.87)
- Contextual Squeeze-and-Excitation for Efficient Few-Shot Image Classification (NeurIPS 2022 · similarity 0.86)
- ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases (ICML 2021 · similarity 0.86)
- Self-Attention-Based Contextual Modulation Improves Neural System Identification (ICLR 2025 · similarity 0.84)
- MCA: Moment Channel Attention Networks (AAAI 2024 · similarity 0.84)
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.