Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Randomized Ensembles for Metric Learning
Hong Xuan, Richard Souvenir, Robert Pless
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 BIER with randomized meta-class ensemble DREML (R,12,48) | embedding model · CUB200 | Recall@1 (%): 55.3 -> 63.9 | Improved |
| replace BIER with randomized meta-class ensemble DREML (R,12,48) | embedding model · CAR196 | Recall@1 (%): 78.0 -> 86.0 | Improved |
| replace BIER with DREML (R,192,48) | embedding model · In-Shop Clothes | Recall@1 (%): 76.9 -> 78.4 | Improved |
| replace BIER with DREML (R,192,12) | embedding model · PKU VehicleID (large test set) | Recall@1 (%): 76.0 -> 83.1 | Improved |
| use ResNet-18 instead of Inception V1 backbone in DREML | DREML (12 meta-classes, 48 embeddings) · CUB200 | Recall@1 (%): 58.9 -> 63.9 | Improved |
| use ResNet-18 instead of Inception V1 backbone in DREML | DREML (12 meta-classes, 48 embeddings) · CAR196 | Recall@1 (%): 84.2 -> 86.0 | Improved |
| use ResNet-18 instead of Inception V1 backbone in DREML | DREML (12 meta-classes, 48 embeddings) · CAR196 | NMI: 71.0 -> 76.4 | Improved |
| increase number of embeddings L in the ensemble (ablation) | DREML · CAR196 | Recall@1 (%) | Improved |
| use extreme numbers of meta-classes D (very few or very many) (ablation) | DREML · CAR196 | Recall@1 (%) | Hurt |
| increase ensemble size to 4 and 24 networks (ablation) | DREML embeddings of unseen classes · CAR196 | dot-product separation of same vs different classes | Improved |
Closest papers by meaning
- Deep Metric Learning with BIER: Boosting Independent Embeddings Robustly (TPAMI 2018 · similarity 0.91)
- BIER - Boosting Independent Embeddings Robustly (ICCV 2017 · similarity 0.90)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.89)
- Attention-based Ensemble for Deep Metric Learning (ECCV 2018 · similarity 0.88)
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.88)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.85)
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.
