Research map / Label-efficient and robust learning / Recognition and long-tail
Divide and Conquer the Embedding Space for Metric Learning
Check 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 |
|---|---|---|---|
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · Stanford Online Products | Recall@1: 72.7 -> 75.9 | Helped |
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · CARS196 | Recall@1: 79.6 -> 84.6; DREML 86.0 | Mixed |
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · CUB200-2011 | Recall@1: 63.6 -> 65.9 | Helped |
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · In-shop Clothes | Recall@1: 82.6 -> 85.7 | Helped |
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · PKU VehicleID large test set | Recall@1: 79.2 -> 82.9 | Helped |
| Divide the embedding space and data into cluster-assigned learners | Margin metric learning · Market-1501 | mAP: 87.9 -> 88.8 | Helped |
| Add embedding-space K-means clustering and split the embedding into learner subspaces (ablation) | Margin metric learning on Stanford Online Products · Stanford Online Products | Recall@1: 75.0 -> 75.9 | Helped |
| Replace embedding-space clustering with random data partitioning (ablation) | Divide-and-conquer metric learning · Stanford Online Products | Recall@1: 75.9 -> 73.2 | Hurt |
| Replace embedding-space K-means with manually grouped ground-truth labels (ablation) | Divide-and-conquer metric learning · Stanford Online Products | Recall@1: 75.9 -> 74.5 | Hurt |
| Use K-means cluster-based sampling without splitting the embedding into learners (ablation) | Margin metric learning · Stanford Online Products | Recall@1: 72.7 -> 75.0 | Helped |
| Concatenate the 8 cluster-assigned learner embeddings rather than use one baseline embedding (ablation) | Stanford Online Products · Stanford Online Products | Recall@1: 72.7 -> 75.9 | Helped |
| Train eight complementary, cluster-assigned learners (ablation) | Stanford Online Products · Stanford Online Products | Correlation coefficient: 0.0602 -> 0.0498 | Helped |
Closest papers by meaning
- Deep Metric Learning via Lifted Structured Feature Embedding (CVPR 2016 · similarity 0.91)
- Deep Metric Learning via Facility Location (CVPR 2017 · similarity 0.89)
- Deep Compositional Metric Learning (CVPR 2021 · similarity 0.89)
- Learning Intra-Batch Connections for Deep Metric Learning (ICML 2021 · similarity 0.89)
- Deep Metric Learning Beyond Binary Supervision (CVPR 2019 · similarity 0.89)
- Cross-Image-Attention for Conditional Embeddings in Deep Metric Learning (CVPR 2023 · 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.