Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Metric Learning with Hierarchical Triplet Loss
Weifeng Ge
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 anchor-neighbor sampling (ablation) | triplet loss with semi-hard mining · In-Shop Clothes Retrieval | Recall@1 (%): 62.3 -> 75.3 | Improved |
| add anchor-neighbor sampling (ablation) | triplet loss with semi-hard mining · CUB-200-2011 | Recall@1 (%): 55.9 -> 56.4 | Improved |
| add dynamic violate margin with flat class tree (ablation) | triplet loss with A-N sampling · In-Shop Clothes Retrieval | Recall@1 (%): 75.3 -> 78.9 | Improved |
| use hierarchical tree of depth 16 instead of flat tree (ablation) | HTL · In-Shop Clothes Retrieval | Recall@1 (%): 78.9 -> 80.9 | Improved |
| increase tree depth from 16 to 32 (ablation) | HTL · In-Shop Clothes Retrieval | Recall@1 (%): 80.9 -> 79.3 | Hurt |
| use tree depth 16 instead of depth 8 (ablation) | HTL · In-Shop Clothes Retrieval | Recall@1 (%): 78.7 -> 80.9 | Improved |
| add hierarchical tree (depth 16) over flat tree (ablation) | HTL · CUB-200-2011 | Recall@1 (%): 56.0 -> 57.1 | Improved |
| add A-N sampling and dynamic violate margin (ablation) | HDC contrastive loss · In-Shop Clothes Retrieval | Recall@1 (%): 62.1 -> 69.4 | Improved |
| add A-N sampling and dynamic violate margin (ablation) | HDC contrastive loss · CUB-200-2011 | Recall@1 (%): 53.6 -> 54.1 | Improved |
| use hard negative mining instead of random sampling (ablation) | triplet loss · In-Shop Clothes Retrieval | Recall@1 (%): 59.3 -> 60.1 | No clear effect |
| use semi-hard instead of hard negative mining (ablation) | triplet loss · In-Shop Clothes Retrieval | Recall@1 (%): 60.1 -> 62.3 | Improved |
| replace triplet loss with HTL | triplet loss baseline · Cars-196 | Recall@1 (%): 79.2 -> 81.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SphereFace — SphereFace: Deep Hypersphere Embedding for Face Recognition (CVPR 2017) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Correcting the Triplet Selection Bias for Triplet Loss (ECCV 2018 · similarity 0.91)
- Hard negative examples are hard, but useful (ECCV 2020 · similarity 0.90)
- A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning (CVPR 2019 · similarity 0.89)
- Improved Deep Metric Learning with Multi-class N-pair Loss Objective (NeurIPS 2016 · similarity 0.88)
- Sharing Matters for Generalization in Deep Metric Learning (TPAMI 2020 · similarity 0.88)
- Triplet-Center Loss for Multi-View 3D Object Retrieval (CVPR 2018 · 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.
