Research map / Architectures and efficiency / CNN design
Global Gated Mixture of Second-order Pooling for Improving Deep Convolutional Neural 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 gated mixture of 16 GAP component models (top-8) | ResNet-18-512d · Downsampled ImageNet-1K val | Top-1 error: 49.08 -> 42.37 | Helped |
| replace averaging of 16 SOPs with sparse gated mixture (ablation) | Ave-SOP-16 · Downsampled ImageNet-1K val | Top-1 error: 40.28 -> 38.21 | Helped |
| replace SR-SOP with parametric SR-SOP (ablation) | SR-SOP · Downsampled ImageNet-1K val | Top-1 error: 40.32 -> 40.01 | Helped |
| use gated mixture of parametric SOPs instead of single parametric SR-SOP (ablation) | Parametric SR-SOP · Downsampled ImageNet-1K val | Top-1 error: 40.01 -> 38.21 | Helped |
| replace gating with direct ensemble averaging of 16 GAP CMs (ablation) | GM-GAP-16-8 · Downsampled ImageNet-1K val | Top-1 error: 42.37 -> 47.44 | Hurt |
| add balance loss (alpha=100 vs alpha=0) (ablation) | GM-GAP-16-4 · Downsampled ImageNet-1K val | Top-1 error | Helped |
| increase K from 8 to 12 or 16 with N=16 (ablation) | GM-GAP · Downsampled ImageNet-1K val | Top-1 error | Hurt |
| increase number of CMs N from 16 to 128 (K=32) (ablation) | GM-GAP · Downsampled ImageNet-1K val | Top-1 error: 42.37 -> 42.52 | No clear effect |
| replace GAP with GM-SOP on ResNet-50 | ResNet-50 · Downsampled ImageNet-1K val | Top-1 error: 43.28 -> 35.73 | Helped |
| replace one-layer SG-MoE with GM-GAP | ResNet-18 · Downsampled ImageNet-1K val | Top-1 error: 46.80 -> 42.37 | Helped |
| add GM-SOP on WRN-36-2 | WRN-36-2 · Downsampled ImageNet-1K val | Top-1 error: 39.55 -> 32.33 | Helped |
| replace SR-SOP with GM-SOP | ResNet-18 · Places365 val (100x100) | Top-1 error: 48.11 -> 47.18 | Helped |
Closest papers by meaning
- Global Second-Order Pooling Convolutional Networks (CVPR 2019 · similarity 0.91)
- What Deep CNNs Benefit From Global Covariance Pooling: An Optimization Perspective (CVPR 2020 · similarity 0.86)
- Gaussian-Based Pooling for Convolutional Neural Networks (NeurIPS 2019 · similarity 0.86)
- S3Pool: Pooling With Stochastic Spatial Sampling (CVPR 2017 · similarity 0.85)
- Detail-Preserving Pooling in Deep Networks (CVPR 2018 · similarity 0.85)
- Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure (ICLR 2025 · 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.