Research map / Label-efficient and robust learning / Recognition and long-tail
Introspective Deep Metric Learning
Chengkun Wang, Wenzhao Zheng, Zheng Dong Zhu, Jie Zhou, Jiwen Lu
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 |
|---|---|---|---|
| add IDML to ProxyAnchor loss | ProxyAnchor · CUB-200-2011 | Recall@1: 69.0 -> 70.7 | Improved |
| add IDML to ProxyAnchor loss | ProxyAnchor · Cars196 | Recall@1: 87.3 -> 90.6 | Improved |
| add IDML to ProxyAnchor loss | ProxyAnchor · Stanford Online Products | Recall@1: 79.5 -> 81.5 | Improved |
| add IDML to triplet loss with semi-hard sampling | Triplet-SH · Cars196 | Recall@1: 70.8 -> 73.7 | Improved |
| add Mixup (without ISM) to ProxyAnchor (ablation) | ProxyAnchor · CUB-200-2011 | Recall@1: 69.0 -> 69.8 | Improved |
| add ISM (without Mixup) to ProxyAnchor (ablation) | ProxyAnchor · Cars196 | Recall@1: 87.3 -> 88.8 | Improved |
| combine ISM with Mixup on top of ISM alone (ablation) | ProxyAnchor · Cars196 | Recall@1: 88.8 -> 90.6 | Improved |
| replace IDML by probabilistic embedding learning (PEL) (ablation) | ProxyAnchor · CUB-200-2011 | Recall@1: 69.0 -> 64.9 | Hurt |
| use IDML instead of vMF distributional modeling (ablation) | ProxyAnchor · CUB-200-2011 | Recall@1: 70.0 -> 70.7 | Improved |
| use ||u1+u2|| instead of ||u1||+||u2|| for uncertainty level (ablation) | IDML-PA · CUB-200-2011 | Recall@1: 70.1 -> 70.7 | Improved |
| use similarity-based ISM (ISM-Sim) instead of distance-based (ISM-Dis) (ablation) | IDML-PA · Cars196 | Recall@1: 89.4 -> 90.6 | Improved |
| use learned uncertainty embedding instead of random noise vectors (ablation) | IDML-PA · CUB-200-2011 | Recall@1: 70.0 -> 70.7 | Improved |
Closest papers by meaning
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.89)
- Deep Compositional Metric Learning (CVPR 2021 · similarity 0.88)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.88)
- Large-Scale Distance Metric Learning With Uncertainty (CVPR 2018 · similarity 0.88)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.87)
- Potential Field Based Deep Metric Learning (CVPR 2025 · 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.
