Research map / Architectures and efficiency / CNN design
Global Second-Order Pooling Convolutional Networks
Check 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 |
|---|---|---|---|
| Introduce second-order pooling across network stages | ResNet-50 using GSoP-Net2 · ImageNet-1K | top-1 error: 22.14 -> 21.19 | Helped |
| Use second-order pooling across network stages with global average pooling at the end | ResNet-50 using GSoP-Net1 · ImageNet-1K | top-1 error: 23.85 -> 22.32 | Helped |
| Use GSoP-Net2 rather than the vanilla backbone | pre-activation ResNet-164 · CIFAR-100 | top-1 error: 24.33 -> 18.58 | Helped |
| Use channel-wise rather than position-wise pooling (ablation) | GSoP-Net1 · ImageNet-1/4K | top-1 error: 18.37 -> 17.42 | Helped |
| Fuse channel-wise and position-wise outputs by averaging (ablation) | GSoP-Net1 · ImageNet-1/4K | top-1 error: 17.42 -> 17.90 | Hurt |
| Fuse channel-wise and position-wise outputs by maximum (ablation) | GSoP-Net1 · ImageNet-1/4K | top-1 error: 17.42 -> 17.48 | No clear effect |
| Fuse channel-wise and position-wise outputs by concatenation (ablation) | GSoP-Net2 · ImageNet-1/4K | top-1 error: 16.68 -> 16.49 | No clear effect |
| Increase channel covariance size from 64×64 to 128×128 (ablation) | GSoP-Net1 · ImageNet-1/4K | top-1 error: 18.00 -> 17.42 | Helped |
| Increase channel covariance size from 128×128 to 256×256 (ablation) | GSoP-Net1 · ImageNet-1/4K | top-1 error: 17.42 -> 17.61 | Hurt |
| Add one channel-wise block at any single residual stage (ablation) | ResNet-26 · ImageNet-1/4K | top-1 error: 19.18 -> 18.45; 19.18 -> 18.72; 19.18 -> 18.85; 19.18 -> 18.33 | Helped |
| Use covariance pooling at the network end rather than global average pooling (ablation) | GSoP-Net2 versus GSoP-Net1 · ImageNet-1/4K | top-1 error: 17.42 -> 16.68 | Helped |
| Introduce second-order pooling in earlier stages in addition to iSQRT-COV at the end (ablation) | ResNet-26 · ImageNet-1/4K | top-1 error: 17.43 -> 16.68 | Helped |
Closest papers by meaning
- Global Gated Mixture of Second-order Pooling for Improving Deep Convolutional Neural Networks (NeurIPS 2018 · similarity 0.91)
- What Deep CNNs Benefit From Global Covariance Pooling: An Optimization Perspective (CVPR 2020 · similarity 0.89)
- Revitalizing SVD for Global Covariance Pooling: Halley’s Method to Overcome Over-Flattening (NeurIPS 2025 · similarity 0.88)
- Deep Global Generalized Gaussian Networks (CVPR 2019 · similarity 0.88)
- G2DeNet: Global Gaussian Distribution Embedding Network and Its Application to Visual Recognition (CVPR 2017 · similarity 0.88)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.86)
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.