Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Generalized Sum Pooling for Metric Learning
Yeti Z. Gürbüz, Ozan Sener, A. Aydin Alatan
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 |
|---|---|---|---|
| Replace global average pooling with GSP | Contrastive (C2) loss · SOP 128D | MAP@R: 41.79 -> 42.84 | Improved |
| Replace global average pooling with GSP | Cars196 with Contrastive loss · Cars196 128D | MAP@R: 17.02 -> 18.31 | Improved |
| Add GSP on top of XBM | XBM+GAP · SOP 128D | MAP@R: 44.55 -> 46.69 | Improved |
| Add zero-shot loss L_ZS to GSP (ablation) | GSP with C2 loss · CUB 128D | MAP@R: 20.85 -> 21.52 | Improved |
| Add GSP to zero-shot loss with GAP (ablation) | GAP + L_ZS with C2 loss · Cars196 128D | MAP@R: 17.67 -> 18.31 | Improved |
| Add L_ZS without GSP to GAP baseline (ablation) | GAP with C2 loss · SOP 128D | MAP@R: 41.79 -> 42.55 | Improved |
| Combine GSP with GeMean pooling | GeMean with C2 loss · CUB 128D | MAP@R: 21.50 -> 21.84 | Improved |
| Increase mu in GSP (ablation) | GSP · Validation sweep | MAP@R | Hurt |
| Use GSP instead of GAP with ProxyNCA++ on In-shop | ProxyNCA++ · In-shop 128D | MAP@R: 54.41 -> 54.68 | Improved |
| Use closed-form gradient instead of automatic differentiation (ablation) | GSP · Computation scaling with iterations k | compute cost | Improved |
Closest papers by meaning
- Global Gated Mixture of Second-order Pooling for Improving Deep Convolutional Neural Networks (NeurIPS 2018 · similarity 0.87)
- Generalizing Pooling Functions in CNNs: Mixed, Gated, and Tree (TPAMI 2017 · similarity 0.86)
- Generalization Bound of Gradient Descent for Non-Convex Metric Learning (NeurIPS 2020 · similarity 0.85)
- Global Second-Order Pooling Convolutional Networks (CVPR 2019 · similarity 0.85)
- Sharing Matters for Generalization in Deep Metric Learning (TPAMI 2020 · similarity 0.85)
- Measuring Generalization with Optimal Transport (NeurIPS 2021 · 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.
