Research map / Label-efficient and robust learning / Semi-supervised learning
SLADE: A Self-Training Framework for Distance Metric Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SwAV self-supervised initialization instead of supervised ImageNet initialization (ablation) | SLADE · CUB-200 and Cars-196 | MAP@R: 29.38 -> 32.79; 31.38 -> 35.54 | Helped |
| fine-tune SwAV on the labeled and unlabeled images without labels before teacher training (ablation) | SLADE · CUB-200 and Cars-196 | MAP@R: 32.79 -> 33.59; 35.54 -> 36.24 | Helped |
| train a student on pseudo-label pairs (ablation) | contrastive teacher · CUB-200 and Cars-196 | MAP@R: 29.29 -> 30.81; 31.73 -> 31.99 | Helped |
| add feature basis learning (ablation) | SLADE student · CUB-200 and Cars-196 | MAP@R: 30.81 -> 32.45; 31.99 -> 35.78 | Helped |
| add high-confidence pair mining using the learned bases (ablation) | SLADE student with feature basis learning · CUB-200 and Cars-196 | MAP@R: 32.45 -> 33.59; 35.78 -> 36.24 | Helped |
| replace similarity distribution loss with local pairwise cross-entropy (ablation) | SLADE feature basis learning · CUB-200 | MAP@R: 33.59 -> 32.69 | Hurt |
| replace similarity distribution loss with global cross-entropy (ablation) | SLADE feature basis learning · CUB-200 | MAP@R: 33.59 -> 32.23 | Hurt |
| vary the number of k-means clusters (ablation) | SLADE on CUB-200 with NABirds unlabeled data · CUB-200 | MAP@R: 31.83 (k=100); 32.61 (k=200); 32.81 (k=300); 33.59 (k=400); 33.26 (k=500) | Mixed |
| use SLADE rather than the multi-similarity teacher alone | multi-similarity retrieval model · CUB-200 and Cars-196 | P@1: 70.81 -> 74.09; 89.33 -> 91.53 | Helped |
| use SLADE with multi-similarity loss instead of Proxy-Anchor | retrieval model · In-shop | Recall@10: 98.1 -> 98.6 | Helped |
Closest papers by meaning
- Self-Taught Metric Learning Without Labels (CVPR 2022 · similarity 0.91)
- Deep Metric Learning with Self-Supervised Ranking (AAAI 2021 · similarity 0.88)
- Neighborhood-Adaptive Structure Augmented Metric Learning (AAAI 2022 · similarity 0.87)
- Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization (ICML 2023 · similarity 0.87)
- Parameter Free Large Margin Nearest Neighbor for Distance Metric Learning (AAAI 2017 · similarity 0.87)
- Embedding Transfer With Label Relaxation for Improved Metric Learning (CVPR 2021 · 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.