Research map / Label-efficient and robust learning / Recognition and long-tail
Ranked List Loss for Deep 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 |
|---|---|---|---|
| Use RLL instead of the strongest listed baseline, Spectral Clust | GoogLeNet V2 embeddings · CARS196 | Recall@1 (%): 73.1 -> 74.0 | Helped |
| Use RLL instead of the strongest listed baseline, Spectral Clust | GoogLeNet V2 embeddings · CUB-200-2011 | Recall@1 (%): 53.2 -> 57.4 | Helped |
| Use RLL instead of the strongest listed baseline, Proxy NCA | GoogLeNet V2 embeddings · SOP | Recall@1 (%): 73.7 -> 76.1 | Helped |
| Concatenate low-, mid-, and high-level embeddings (ablation) | RLL · SOP | Recall@1 (%): 76.1 -> 79.8 | Helped |
| Set the positive-negative margin m to 0 instead of 0.4 (ablation) | RLL · SOP | Recall@1 (%): 79.8 -> 76.1 | Hurt |
| Use negative-example weighting with T = 10 rather than no weighting (ablation) | RLL · SOP | Recall@1 (%): 78.8 -> 79.8 | Helped |
| Vary the negative-example weighting temperature T from 0 to 20 (ablation) | RLL · SOP | Recall@1 (%): 78.8; 79.1; 79.8; 79.3; 78.6 | Mixed |
| Use margin m values from 0.2 to 1.2 (ablation) | RLL · SOP | Recall@1 (%): 79.0; 79.8; 79.2; 79.1 | No clear effect |
| Change α from 1.2 to 1.0 or 1.4 (ablation) | RLL · SOP | Recall@1 (%): 79.8 -> 78.7; 76.2 | Hurt |
| Use only one embedding level instead of concatenating L, M, and H (ablation) | RLL · SOP | Recall@1 (%): 79.8 -> 76.1; 76.9; 76.1 | Hurt |
| Vary batch size from 120 to 195 (ablation) | RLL · SOP | Recall@1 (%): 79.2; 79.5; 79.7; 79.8; 79.8 | No clear effect |
Closest papers by meaning
- Deep Metric Learning with Self-Supervised Ranking (AAAI 2021 · similarity 0.90)
- Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning (CVPR 2019 · similarity 0.89)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.89)
- Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization (ICML 2023 · similarity 0.88)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · similarity 0.87)
- Deep Disentangled Metric Learning (AAAI 2025 · 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.