Research map / Label-efficient and robust learning / Recognition and long-tail
Generalized Orderless Pooling Performs Implicit Salient Matching
Marcel Simon, Yang Gao, Trevor Darrell, Joachim Denzler, Erik Rodner
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 |
|---|---|---|---|
| learn alpha (generalized pooling) instead of fixing bilinear pooling (alpha=2) | VGG16 fine-tuned · CUB200-2011 | top-1 accuracy (%): 84.1% -> 85.3% | Improved |
| learn alpha instead of fixing bilinear pooling (alpha=2) | VGG16 fine-tuned · FGVC-Aircraft | top-1 accuracy (%): 84.1% -> 85.5% | Improved |
| learn alpha pooling instead of best previous method on actions | VGG16 fine-tuned · Stanford 40 actions | top-1 accuracy (%): 81.7% -> 86.0% | Improved |
| replace fully-connected layers with alpha-pooling when training from scratch | VGG-M with batch norm · ImageNet ILSVRC 2012 | top-1 validation accuracy | Improved |
| increase alpha from 1 (average pooling) to 3 (ablation) | VGG16 without fine-tuning · CUB200-2011 sample bird image | contribution of most relevant training region (%): 11.2% -> 23.2% | Improved |
| increase alpha from 1 to 2 (ablation) | VGG16 without fine-tuning · CUB200-2011 | bird head contribution to decision (%): 9% -> 42% | Improved |
| increase alpha to focus on fewer feature matches (ablation) | VGG16 without fine-tuning · generic datasets (MIT Scenes 67) | normalized accuracy | Hurt |
| increase alpha (best alpha rises from 1.3 to 2.5 across datasets) (ablation) | VGG16 without fine-tuning · MIT Scenes 67, 40 actions, CUB200-2011 | normalized accuracy: 1.3, 1.5, 2.5 | Mixed |
| add fine-tuning with alpha fixed to 2 (ablation) | VGG16 · CUB200-2011 | part contribution to decision | Improved |
Closest papers by meaning
- Generalizing Pooling Functions in CNNs: Mixed, Gated, and Tree (TPAMI 2017 · similarity 0.89)
- Kernel Pooling for Convolutional Neural Networks (CVPR 2017 · similarity 0.88)
- Compact Bilinear Pooling (CVPR 2016 · similarity 0.88)
- Hierarchical Bilinear Pooling for Fine-Grained Visual Recognition (ECCV 2018 · similarity 0.88)
- Learning Compact Discriminant Representation via Low-Rank Bilinear Pooling (TPAMI 2025 · similarity 0.87)
- Statistically-motivated Second-order Pooling (ECCV 2018 · similarity 0.87)
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.
