Research map / Label-efficient and robust learning / Recognition and long-tail
SoftTriple Loss: Deep Metric Learning Without Triplet Sampling
Qi Qian, Lei Shang, Baigui Sun, Juhua Hu, Hao Li, Rong Jin
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 |
|---|---|---|---|
| use multiple centers per class (K=10) instead of one center (SoftMax_norm) (ablation) | SoftTriple · CUB-2011 (64-d) | Recall@1: 57.8 -> 60.1 | Improved |
| use multiple centers per class (K=10) instead of one center (SoftMax_norm) (ablation) | SoftTriple · CUB-2011 (large embeddings) | Recall@1: 64.2 -> 65.4 | Improved |
| use multiple centers per class (K=10) instead of one center (SoftMax_norm) (ablation) | SoftTriple · Cars196 (64-d) | Recall@1: 76.8 -> 78.6 | Improved |
| use multiple centers per class (K=10) instead of one center (SoftMax_norm) (ablation) | SoftTriple · Cars196 (large embeddings) | Recall@1: 83.2 -> 84.5 | Improved |
| increase centers from 1 to 2 (K=2, no regularizer) (ablation) | SoftTriple · SOP (64-d) | Recall@1: 75.9 -> 76.3 | Improved |
| use multiple centers instead of one center (ablation) | SoftTriple · SOP (large embeddings) | Recall@1: 78.0 -> 78.3 | Improved |
| replace ProxyNCA with normalized SoftMax loss | SoftMax_norm · Cars196 (64-d) | Recall@1: 73.2 -> 76.8 | Improved |
| replace ProxyNCA with normalized SoftMax loss | SoftMax_norm · SOP (64-d) | Recall@1: 73.7 -> 75.9 | Improved |
| increase number of centers from 1 to 10 (ablation) | SoftTriple with regularizer · CUB-2011 | Recall@1 | Improved |
| increase number of centers beyond 10 with the regularizer (ablation) | SoftTriple with regularizer · CUB-2011 | Recall@1 | No clear effect |
| add the L2,1 center-merging regularizer when centers are over-parameterized (ablation) | SoftTriple · CUB-2011 (K=20) | Recall@1 | Improved |
| add the regularizer to shrink similar centers (ablation) | SoftTriple · CUB-2011 (K=20) | number of unique centers per class | Improved |
Closest papers by meaning
- A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning (CVPR 2019 · similarity 0.88)
- DAS: Densely-Anchored Sampling for Deep Metric Learning (ECCV 2022 · similarity 0.88)
- OLÉ: Orthogonal Low-Rank Embedding - A Plug and Play Geometric Loss for Deep Learning (CVPR 2018 · similarity 0.87)
- Deep Metric Learning with Hierarchical Triplet Loss (ECCV 2018 · similarity 0.87)
- Ranked List Loss for Deep Metric Learning (CVPR 2019 · similarity 0.87)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 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.
