Research map / Label-efficient and robust learning / Recognition and long-tail
Sharing Matters for Generalization in Deep Metric Learning
Timo Milbich, Karsten Roth, Biagio Brattoli, Björn Ommer
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 shared-feature task with separate encoder | Margin loss ResNet50 · CARS196 | Recall@1: 81.5 -> 87.0 | Improved |
| add shared-feature task | Margin loss ResNet50 · CUB200-2011 | Recall@1: 65.2 -> 68.6 | Improved |
| add shared-feature task | Margin loss ResNet50 · SOP | Recall@1: 76.1 -> 78.2 | Improved |
| add shared-feature task | Semihard triplet loss ResNet50 · CARS196 | Recall@1: 72.7 -> 79.4 | Improved |
| add shared-feature task | N-pairs ResNet50 · CARS196 | Recall@1: 70.6 -> 77.2 | Improved |
| add shared-feature task | ProxyNCA ResNet50 · CARS196 | Recall@1: 80.8 -> 82.7 | Improved |
| add shared task to discriminative baseline with single encoder (ablation) | discr. only · CARS196 | Recall@1: 81.5 -> 83.2 | Improved |
| use separate encoders for discriminative and shared tasks (ablation) | both · CARS196 | Recall@1: 83.2 -> 84.1 | Improved |
| add decorrelation between embeddings (ablation) | both+sep · CARS196 | Recall@1: 84.1 -> 87.0 | Improved |
| increase decorrelation weight too large (ablation) | both+sep+decor · CARS196 | Recall@1: 87.0 -> 82.1 | Hurt |
| train only on shared triplets without discriminative task (ablation) | discr. only · CARS196 | Recall@1: 81.5 -> 36.1 | Hurt |
| add feature standardization before grouping (ablation) | group triplets · CARS196 | Recall@1: 82.3 -> 85.8 | Improved |
Closest papers by meaning
- Deep Compositional Metric Learning (CVPR 2021 · similarity 0.90)
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.90)
- Divide and Conquer the Embedding Space for Metric Learning (CVPR 2019 · similarity 0.89)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.89)
- DarkRank: Accelerating Deep Metric Learning via Cross Sample Similarities Transfer (AAAI 2018 · similarity 0.89)
- Improved Deep Metric Learning with Multi-class N-pair Loss Objective (NeurIPS 2016 · similarity 0.88)
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.
