Research map / Label-efficient and robust learning / Knowledge distillation
Asymmetric Metric Learning for Knowledge Transfer
Mateusz Budnik, Yannis Avrithis
Check 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 asymmetric metric learning with contrastive loss rather than symmetric student-only contrastive training | MobileNetV2 · R-Oxford5k medium, symmetric; VGG16 teacher | mAP: 53.6 -> 57.3 | Improved |
| Add the teacher's anchor representation as a positive for the student anchor (ablation) | Contrastive asymmetric training · R-Oxford5k medium, symmetric; MobileNetV2, VGG16 teacher | mAP: 57.3 -> 57.3 | Improved |
| Add negatives to contrastive asymmetric training that uses only positive pairs (ablation) | Contrastive asymmetric training · R-Oxford5k medium, symmetric; MobileNetV2, VGG16 teacher | mAP: 55.9 -> 57.3 | Improved |
| Combine self-positive regression with labeled contrastive metric learning (ablation) | MobileNetV2 · R-Oxford5k medium, asymmetric; VGG16 teacher | mAP: 48.0 -> 42.9 | Hurt |
| Use plain feature regression instead of relational distillation | MobileNetV2, VGG16 teacher · R-Oxford5k medium, asymmetric | mAP: 2.0 -> 48.0 | Improved |
| Use plain feature regression instead of DarkRank | MobileNetV2, VGG16 teacher · R-Oxford5k medium, asymmetric | mAP: 1.7 -> 48.0 | Improved |
| Use Contr+ rather than plain contrastive training (ablation) | MobileNetV2, ResNet101 teacher · R-Oxford5k medium, asymmetric | mAP: 32.3 -> 47.1 | Improved |
| Use asymmetric metric learning rather than student-only training | EfficientNet-B3, ResNet101 teacher · R-Paris6k medium, symmetric | mAP: 75.1 -> 77.1 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DarkRank — DarkRank: Accelerating Deep Metric Learning via Cross Sample Similarities Transfer (AAAI 2018) · better on 1 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
- Self-Taught Metric Learning Without Labels (CVPR 2022 · similarity 0.87)
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · similarity 0.87)
- Deep Asymmetric Metric Learning via Rich Relationship Mining (CVPR 2019 · similarity 0.86)
- Asymmetric Decision-Making in Online Knowledge Distillation: Unifying Consensus and Divergence (ICML 2025 · similarity 0.85)
- Less or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation (ICLR 2024 · similarity 0.85)
- Towards Improved Proxy-Based Deep Metric Learning via Data-Augmented Domain Adaptation (AAAI 2024 · similarity 0.85)
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.
