Research map / Label-efficient and robust learning / Recognition and long-tail
Smart Mining for Deep Metric Learning
Ben Harwood, Vijay Kumar B G, Gustavo Carneiro, Ian Reid, Tom Drummond
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 semi-hard stochastic mining with FANNG smart mining | triplet loss · CUB-200-2011 | Recall@1: 42.59 -> 45.90 | Improved |
| replace semi-hard stochastic mining with FANNG smart mining | triplet loss · Cars196 | Recall@1: 51.54 -> 56.11 | Improved |
| add global loss (ablation) | Triplet + FANNG · CUB-200-2011 | Recall@1: 45.90 -> 49.44 | Improved |
| add global loss (ablation) | Triplet + FANNG · Cars196 | Recall@1: 56.11 -> 64.20 | Improved |
| add global loss (ablation) | Triplet + FANNG · CUB-200-2011 | NMI: 58.10 -> 60.09 | Improved |
| add adaptive controller (ablation) | Triplet + FANNG + Global · CUB-200-2011 | Recall@8: 82.85 -> 83.31 | Improved |
| add adaptive controller (ablation) | Triplet + FANNG + Global · Cars196 | Recall@8: 88.94 -> 90.19 | Improved |
| add adaptive controller (ablation) | Triplet + FANNG + Global · CUB-200-2011 | NMI: 60.09 -> 59.90 | No clear effect |
| add adaptive controller (ablation) | Triplet + FANNG + Global · Cars196 | convergence speed | Improved |
| use FANNG mined triplets in place of random triplets (global loss kept) versus Triplet + Global | Triplet + Global · Cars196 | Recall@1: 61.41 -> 64.20 | Improved |
| use smaller initial kappa (kappa in {1,4,16,64}) (ablation) | Triplet + FANNG + Global · CUB-200-2011 | Recall@K after 20 epochs | Hurt |
| use a low percentage (10-30%) of mined triplets (ablation) | Triplet + FANNG + Adaptive · CUB-200-2011 | training error control / convergence | Hurt |
Closest papers by meaning
- Deep Metric Learning by Online Soft Mining and Class-Aware Attention (AAAI 2019 · similarity 0.88)
- LoOp: Looking for Optimal Hard Negative Embeddings for Deep Metric Learning (ICCV 2021 · similarity 0.87)
- Stochastic Class-Based Hard Example Mining for Deep Metric Learning (CVPR 2019 · similarity 0.87)
- Mathematical Justification of Hard Negative Mining via Isometric Approximation Theorem (ICLR 2024 · similarity 0.87)
- A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning (CVPR 2019 · similarity 0.86)
- Hardness-Aware Deep Metric Learning (CVPR 2019 · similarity 0.86)
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.
