Research map / Label-efficient and robust learning / Recognition and long-tail
Improving Deep Metric Learning by Divide and Conquer
Artsiom Sanakoyeu, Pingchuan Ma, Vadim Tschernezki, Björn Ommer
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 |
|---|---|---|---|
| use progressive division instead of directly dividing into K_max clusters (ablation) | Ours (fix) · SOP | Recall@1: 78.67 -> 79.54 | Improved |
| cluster in the learned embedding space instead of random splits (ablation) | Ours (fix) · SOP | Recall@1: 71.99 -> 79.54 | Improved |
| cluster in the learned embedding space instead of random splits (ablation) | Ours (fix) · CARS196 | Recall@1: 85.02 -> 86.46 | Improved |
| cluster in the learned embedding space instead of random splits (ablation) | Ours (fix) · CUB200-2011 | Recall@1: 66.96 -> 68.43 | Improved |
| adaptive clustering instead of ground-truth super-category grouping (ablation) | Ours (fix) · SOP | Recall@1: 75.99 -> 79.54 | Improved |
| adaptive clustering instead of ground-truth super-category grouping (ablation) | Ours (fix) · CUB200-2011 | Recall@1: 65.02 -> 68.43 | Improved |
| add separate subspaces per cluster (ablation) | Ours (fix) · SOP | Recall@1: 79.23 -> 79.54 | Improved |
| add separate subspaces per cluster (ablation) | Ours (fix) · CARS196 | Recall@1: 83.90 -> 86.46 | Improved |
| add separate subspaces per cluster (ablation) | Ours (fix) · In-shop Clothes | Recall@1: 89.17 -> 90.15 | Improved |
| use learnable masks instead of fixed orthogonal masks (ablation) | Ours (fix) · CARS196 | Recall@1: 86.46 -> 87.76 | Improved |
| use learnable masks instead of fixed orthogonal masks (ablation) | Ours (fix) · CUB200-2011 | Recall@1: 68.43 -> 68.16 | No clear effect |
| increase K_max from 1 to 32 (ablation) | Margin loss · SOP | Recall@1: 77.11 -> 79.54 | Improved |
Closest papers by meaning
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.96)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.92)
- Deep Metric Learning via Lifted Structured Feature Embedding (CVPR 2016 · similarity 0.91)
- DiVA: Diverse Visual Feature Aggregation for Deep Metric Learning (ECCV 2020 · similarity 0.91)
- Integrating Language Guidance Into Vision-Based Deep Metric Learning (CVPR 2022 · similarity 0.90)
- Potential Field Based Deep Metric Learning (CVPR 2025 · similarity 0.89)
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.
