Research map / Architectures and efficiency / Quantization
Statistically-motivated Second-order Pooling
Kaicheng Yu, Mathieu Salzmann
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 parametric vectorization (PV) layer p=2048 in place of full vectorization (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 75.97 -> 75.32 | No clear effect |
| add square-root transformation after the PV layer (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 75.32 -> 76.32 | Improved |
| add learned scale and bias after the square-root (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 76.32 -> 78.00 | Improved |
| add square-root transformation with scale and bias to the PV layer (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 76.14 -> 78.00 | Improved |
| use logarithm instead of square-root transformation (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 78.00 -> 77.86 | No clear effect |
| use cubic-root instead of square-root transformation (ablation) | SMSO (VGG-D) · MINC-2500 | top-1 accuracy: 78.00 -> 77.17 | No clear effect |
| vary PV dimension p from 16 to 8192 (ablation) | SMSO (VGG-D) · DTD, MIT, MINC | top-1 accuracy | No clear effect |
| use SMSO p=2048 instead of plain fine-tuned VGG-D | VGG-D · MINC-2500 | top-1 accuracy: 73.01 -> 78.00 | Improved |
| use SMSO p=2048 instead of MPN | VGG-D · MINC-2500 | top-1 accuracy: 76.24 -> 78.00 | Improved |
| use SMSO p=2048 instead of plain ResNet-50 | ResNet-50 · MIT-Indoor | top-1 accuracy: 76.45 -> 79.68 | Improved |
| use MPN second-order pooling instead of plain ResNet-50 | ResNet-50 · MIT-Indoor | top-1 accuracy: 76.45 -> 72.12 | Hurt |
| use learnable PV projection instead of random fixed projection (CBP-TS) | SMSO p=2048 (VGG-D) · MINC-2500 | top-1 accuracy: 73.28 -> 78.00 | Improved |
Closest papers by meaning
- Global Second-Order Pooling Convolutional Networks (CVPR 2019 · similarity 0.88)
- Second-Order Pooling for Graph Neural Networks (TPAMI 2020 · similarity 0.87)
- MoNet: Moments Embedding Network (CVPR 2018 · similarity 0.87)
- Global Gated Mixture of Second-order Pooling for Improving Deep Convolutional Neural Networks (NeurIPS 2018 · similarity 0.87)
- On the Parameterization of Second-Order Optimization Effective towards the Infinite Width (ICLR 2024 · similarity 0.86)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · 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.
