Research map / Label-efficient and robust learning / Recognition and long-tail
DAS: Densely-Anchored Sampling for Deep Metric Learning
Lizhao Liu, Shangxin Huang, Zhuangwei Zhuang, Ran Yang, Mingkui Tan, Yaowei Wang
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 DAS to ResNet50-512 margin-style MS | MS · CUB | Recall@1 (%): 66.46 -> 69.19 | Improved |
| add DAS to ResNet50-512 MS | MS · CARS | Recall@1 (%): 83.99 -> 87.84 | Improved |
| add DAS to triplet loss with distance weighted sampling | Triplet [D] · CARS | Recall@1 (%): 78.86 -> 82.63 | Improved |
| add DAS to contrastive loss with distance sampling | Contrastive [D] · CARS | Recall@1 (%): 76.03 -> 80.74 | Improved |
| add DAS to whole-batch loss GenLifted | GenLifted · CARS | Recall@1 (%): 72.45 -> 73.55 | Improved |
| add DAS to triplet loss with random sampling | Triplet random sampling · CARS | Recall@1 (%): 74.21 -> 76.79 | Improved |
| add DAS to contrastive loss with random sampling | Contrastive random sampling · CARS | Recall@1 (%): 42.44 -> 50.79 | Improved |
| add DFS module (ablation) | Triplet [D] + DAS · CARS | Recall@1 (%): 78.86 -> 81.28 | Improved |
| add MTS module (ablation) | Triplet [D] + DAS · CARS | NMI: 65.85 -> 67.81 | Improved |
| add MTS on top of DFS (ablation) | Triplet [D] + DAS · CARS | Recall@1 (%): 81.28 -> 82.63 | Improved |
| increase K in DFS beyond 4 (ablation) | DFS with margin loss · SOP | Recall@1 (%): 79.14 -> 78.09 | Hurt |
| increase bank capacity Z in MTS (ablation) | MTS with margin loss · CARS | Recall@1 (%): 82.35 -> 82.75 | Improved |
Closest papers by meaning
- Improving Deep Metric Learning by Divide and Conquer (TPAMI 2021 · similarity 0.90)
- Revisiting Training Strategies and Generalization Performance in Deep Metric Learning (ICML 2020 · similarity 0.88)
- Sampling Matters in Deep Embedding Learning (ICCV 2017 · similarity 0.88)
- SoftTriple Loss: Deep Metric Learning Without Triplet Sampling (ICCV 2019 · similarity 0.88)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.88)
- Ranked List Loss for Deep Metric Learning (CVPR 2019 · similarity 0.87)
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.
