Research map / Label-efficient and robust learning / Recognition and long-tail
Hard negative examples are hard, but useful
Hong Xuan, Abby Stylianou, Xiaotong Liu, Robert Pless
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 hard negative mining with standard triplet loss (HN) instead of SCT (ablation) | triplet loss with hard negative mining · CUB, CAR, Hotels-50K (high hard-triplet ratio early in training) | training convergence | Hurt |
| add selectively contrastive triplet loss (SCT) in place of semi-hard negative mining (SHN) | triplet loss with hard negative mining · Hotels-50K | Recall@1 | Improved |
| add SCT in place of SHN | triplet loss with hard negative mining · CUB, CAR, SOP, In-shop Cloth | Recall@1 | Improved |
| add SCT instead of SHN, which only weakly separates hard negatives | triplet loss with hard negative mining · CUB and CAR triplet diagram | hard negative separation | Improved |
| add SCT instead of SHN for unseen-class generalization | triplet loss with hard negative mining · CAR (unseen test classes) | anchor-positive similarity on test classes | Improved |
| add SCT loss at the margin-based formulation instead of NCA-based (ablation) | triplet loss · numerical simulation of gradient update | problematic region behavior | No clear effect |
Closest papers by meaning
- Correcting the Triplet Selection Bias for Triplet Loss (ECCV 2018 · similarity 0.90)
- Deep Metric Learning with Hierarchical Triplet Loss (ECCV 2018 · similarity 0.90)
- Rethinking Preventing Class-Collapsing in Metric Learning With Margin-Based Losses (ICCV 2021 · similarity 0.90)
- No Fuss Distance Metric Learning Using Proxies (ICCV 2017 · similarity 0.88)
- A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning (CVPR 2019 · similarity 0.88)
- Mathematical Justification of Hard Negative Mining via Isometric Approximation Theorem (ICLR 2024 · 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.
