Research map / Architectures and efficiency / RNNs and normalization
Deep Factorized Metric Learning
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 |
|---|---|---|---|
| Factorize the DeiT-Small backbone and route samples to sub-blocks | ProxyAnchor metric learning · CUB-200-2011 | Recall@1: 77.5 -> 79.1 | Helped |
| Factorize the DeiT-Small backbone and route samples to sub-blocks | ProxyAnchor metric learning · Cars196 | Recall@1: 87.7 -> 89.5 | Helped |
| Factorize the DeiT-Small backbone and route samples to sub-blocks | ProxyAnchor metric learning · Stanford Online Products | Recall@1: 82.7 -> 84.2 | Helped |
| Factorize only the MLP blocks | ViT-B image classification · ImageNet-1K | Top-1 accuracy: 81.8 -> 82.0 (+0.2) | Helped |
| Use 1 sub-block instead of 3 (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 79.1 -> 77.5 | Hurt |
| Use 2 sub-blocks instead of 3 (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 79.1 -> 78.8 | Hurt |
| Use 6 sub-blocks instead of 3 (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 79.1 -> 78.4 | Hurt |
| Add the complete-output discriminative loss Jdis (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 70.6 -> 79.1 | Helped |
| Add the factorized-embedding discriminative loss Jd (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 77.5 -> 79.1 | Helped |
| Add the factorization loss Jf (ablation) | DFML-PA · CUB-200-2011 | Recall@1: 78.5 -> 79.1 | Helped |
| Replace hard routing with soft gating (ablation) | ProxyAnchor metric learning · CUB-200-2011 | Recall@1: 77.5 -> 77.4 | Hurt |
| Factorize attention and MLP blocks instead of using the baseline network (ablation) | ViT-B image classification, 300 epochs · ImageNet-1K | Top-1 accuracy: 81.8 -> 81.4 (-0.4) | Hurt |
Closest papers by meaning
- Deep Compositional Metric Learning (CVPR 2021 · similarity 0.88)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.88)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.88)
- Deep Disentangled Metric Learning (AAAI 2025 · similarity 0.86)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.85)
- Potential Field Based Deep Metric Learning (CVPR 2025 · 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.