Research map / Label-efficient and robust learning / Recognition and long-tail
Dynamic Metric Learning with Cross-Level Concept Distillation
Wenzhao Zheng, Yuanhui Huang, Borui Zhang, Jie Zhou, Jiwen Lu
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 stop-gradient on lower-level concept targets (ablation) | CLCD · DyML-Vehicle | mAP: 1.3 -> 16.6 | Improved |
| add stop-gradient on lower-level concept targets (ablation) | CLCD · DyML-Vehicle | R@1: 10.2 -> 43.7 | Improved |
| use symmetric pulling instead of asymmetry (ablation) | CLCD · DyML-Vehicle | R@1: 40.1 -> 43.7 | Improved |
| use hierarchical sampling instead of random sampling (ablation) | CLCD · DyML-Vehicle | R@1: 30.6 -> 43.7 | Improved |
| use cross-level distillation instead of intra-level positive pulling (ablation) | CLCD · DyML-Vehicle | R@1: 34.3 -> 43.7 | Improved |
| add negative repulsion at fine level (ablation) | CLCD · DyML-Product fine level | mAP: 13.9 -> 16.2 | Improved |
| add negative repulsion at fine level (ablation) | CLCD · DyML-Product coarse level | R@1: 89.8 -> 86.9 | Hurt |
| add negative repulsion at fine level (ablation) | CLCD · DyML-Product overall | R@1: 59.5 -> 55.4 | Hurt |
| use ImageNet pretrained weights, ICR vs ACR | CLCD-ACR · DyML-Animal | R@1: 57.1 -> 83.0 | Improved |
| ICR instead of ACR with pretrained weights | CLCD · DyML-Animal (pretrained) | R@1: 83.0 -> 83.3 | Improved |
| use CLCD instead of CSL | CLCD-ICR · DyML-Vehicle | R@1: 25.2 -> 43.7 | Improved |
| use positive-attraction-only CLCD-V instead of CosFace | CLCD-V · CUB-200-2011 (concatenated) | P@1: 67.32 -> 67.13 | No clear effect |
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.
- CosFace — CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018) · 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
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · similarity 0.86)
- Dynamic Metric Learning: Towards a Scalable Metric Space To Accommodate Multiple Semantic Scales (CVPR 2021 · similarity 0.85)
- MIC: Mining Interclass Characteristics for Improved Metric Learning (ICCV 2019 · similarity 0.85)
- Wasserstein Distance Rivals Kullback-Leibler Divergence for Knowledge Distillation (NeurIPS 2024 · similarity 0.85)
- Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding (CVPR 2024 · similarity 0.84)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.83)
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.
