Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Compositional 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 learnable compositors to form sub-embedding composites for auxiliary tasks | DCML-MDW · CUB-200-2011 | R@1: 66.2 -> 68.4 | Helped |
| use learnable compositors to form sub-embedding composites for auxiliary tasks | DCML-MDW · Cars196 | R@1: 82.9 -> 85.2 | Helped |
| use learnable compositors to form sub-embedding composites for auxiliary tasks | DCML-MDW · Stanford Online Products | R@1: 78.4 -> 79.8 | Helped |
| add compositors rather than apply subtasks directly to sub-embeddings (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 66.2 -> 68.4 | Helped |
| add the self-reinforced compositor objective (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 67.2 -> 68.4 | Helped |
| add the subtask loss to compositor training (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 67.4 -> 68.4 | Helped |
| add the discriminative metric loss on concatenated sub-embeddings (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 68.2 -> 68.4 | No clear effect |
| use the proposed compositor instead of a direct fully connected layer (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 64.8 -> 68.4 | Helped |
| use the proposed compositor instead of Exp → ReLU (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 67.1 -> 68.4 | Helped |
| retain tanh in the compositor (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | R@1: 67.9 -> 68.4 | No clear effect |
| increase the number of learners beyond four while keeping four compositors (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | retrieval and clustering performance | No clear effect |
| increase the number of compositors (ablation) | DCML on CUB-200-2011 · CUB-200-2011 | retrieval and clustering performance | Helped |
Closest papers by meaning
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.90)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.89)
- Distance Metric Learning with Joint Representation Diversification (ICML 2020 · similarity 0.89)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.88)
- Potential Field Based Deep Metric Learning (CVPR 2025 · similarity 0.87)
- 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.