Research map / Label-efficient and robust learning / Recognition and long-tail
Distance Metric Learning with Joint Representation Diversification
Read the paperCheck 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 JRS regularizer (joint representation similarity) to AMSoftmax | AMSoftmax deep DML · CUB | Recall@1 (%): JRD 50.7 vs MRD 49.4 (Table 2 uses CI-averaged splits) | Helped |
| use joint representations across layers vs marginal DML representation only in JRS (ablation) | JRD regularizer · CUB, CARS, SOP | Recall@1 (%): CUB 50.7 vs 49.4; CARS 61.2 vs 59.8; SOP 79.2 vs 78.8 | Helped |
| include pooling and class-level representations in JRS (vs only pooling+DML) (ablation) | JRD regularizer · CUB, CARS, SOP | Recall@1 (%): CUB 50.7 vs JRD-C 48.6; CARS 61.2 vs 58.5; SOP 79.2 vs 77.7 | Helped |
| replace inner-product JRS with norm-based JMMD or intra-class regularizer JIntra (ablation) | AMSoftmax with JRS · CUB | Recall@1 | Hurt |
| use mixture of Gaussian kernels vs single Gaussian kernel in JRS (ablation) | JRD · CUB, CARS | Recall@1 | Helped |
| Laplace kernel instead of Gaussian in JRS (ablation) | JRD · CUB | Recall@1 (%): 68.1 vs 67.9 | No clear effect |
| polynomial and moment-generating kernels instead of Gaussian in JRS (ablation) | JRD · CUB | Recall@1 (%): 66.1 / 65.2 / 66.1 vs 67.9 | Hurt |
| JRD with JRS regularizer vs prior state of the art on SOP | AMSoftmax deep DML · SOP | Recall@1 (%): 79.2 vs 78.3 (SoftTriple) | Helped |
| JRD with JRS regularizer vs prior state of the art on CUB | AMSoftmax deep DML · CUB | Recall@1 (%): 67.9 vs 65.7 (MS) | Helped |
| JRD vs DeML on CARS | AMSoftmax deep DML · CARS | Recall@1 (%): 84.7 vs 86.3 (DeML) | Hurt |
Closest papers by meaning
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.90)
- Deep Compositional Metric Learning (CVPR 2021 · similarity 0.89)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.89)
- Potential Field Based Deep Metric Learning (CVPR 2025 · similarity 0.86)
- Volume-Aware Distance for Robust Similarity Learning (ICML 2025 · similarity 0.86)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · 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.