Research map / Label-efficient and robust learning / Recognition and long-tail
Moving in the Right Direction: A Regularization 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 |
|---|---|---|---|
| Add direction regularization to MS loss (ablation) | MS loss · CUB-200-2011 | Recall@1 (%): 57.4 -> 59.1 | Helped |
| Add direction regularization to Triplet Loss (ablation) | Triplet Loss · CUB-200-2011 | Recall@1 (%): 51.9 -> 54.2 | Helped |
| Add direction regularization to Proxy-NCA (ablation) | Proxy-NCA · CUB-200-2011 | Recall@1 (%): 49.2 -> 53.8 | Helped |
| Use direction-regularized MS loss with embedding dimension 512 instead of 64 | DR-MS · CUB-200-2011 | Recall@1 (%): 59.1 -> 66.1 | Helped |
| Use direction-regularized MS loss with embedding dimension 512 instead of 64 | DR-MS · Cars-196 | Recall@1 (%): 79.3 -> 85.0 | Helped |
| Set direction regularization weight γ to 0.1 instead of 0.0 (ablation) | DR-MS · CUB-200-2011 | Recall@1 (%): 57.4 -> 58.7 | Helped |
| Set direction regularization weight γ to 0.2 instead of 0.0 (ablation) | DR-MS · CUB-200-2011 | Recall@1 (%): 57.4 -> 59.1 | Helped |
| Set direction regularization weight γ to 0.3 instead of 0.0 (ablation) | DR-MS · CUB-200-2011 | Recall@1 (%): 57.4 -> 60.5 | Helped |
| Increase direction regularization weight γ from 0.3 to 0.4 (ablation) | DR-MS · CUB-200-2011 | Recall@1 (%): 60.5 -> 57.0 | Hurt |
| Use learnable γ instead of static γ = 0.3 (ablation) | DR-MS · CUB-200-2011 | Recall@1 (%): 60.5 -> 59.1 | Hurt |
| Increase training batch size from 80 to 160 (ablation) | DR-MS512 · In-Shop | Recall@1 (%): 87.4 -> 88.3 | Helped |
| Increase training batch size from 160 to 320 (ablation) | DR-MS512 · In-Shop | Recall@1 (%): 88.3 -> 89.5 | Helped |
Closest papers by meaning
- Multi-level Distance Regularization for Deep Metric Learning (AAAI 2021 · similarity 0.88)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.87)
- Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning (CVPR 2019 · similarity 0.87)
- Non-Isotropy Regularization for Proxy-Based Deep Metric Learning (CVPR 2022 · similarity 0.87)
- Deep Metric Learning with Graph Consistency (AAAI 2021 · similarity 0.86)
- Mean Field Theory in Deep Metric Learning (ICLR 2024 · 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.