Research map / Label-efficient and robust learning / Knowledge distillation
ICD-Face: Intra-class Compactness Distillation for Face Recognition
Zhipeng Yu, Jiaheng Liu, Haoyu Qin, Yichao Wu, Kun Hu, Jiayi Tian, Ding Liang
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 intra-class compactness distillation (SDC loss) on top of feature consistency distillation (ablation) | MobileNetV2 student · IJB-C (TAR@FAR=1e-4) | TAR (%): 92.68 -> 93.57 | Improved |
| add intra-class compactness distillation on top of FCD (ablation) | MobileNetV2 student · IJB-B (TAR@FAR=1e-5) | TAR (%): 81.92 -> 85.32 | Improved |
| use feature consistency (L2) distillation instead of classical KD for face recognition | MobileNetV2 student · IJB-C (TAR@FAR=1e-4) | TAR (%): 89.03 -> 92.68 | Improved |
| add ArcFace classification loss to ICD-Face (ICD-Face+) (ablation) | ICD-Face · IJB-C (TAR@FAR=1e-4) | TAR (%): 93.57 -> 93.81 | Improved |
| replace CoupleFace with ICD-Face | face distillation · MegaFace (10^6 distractors) | Rank-1 accuracy (%): 94.27 -> 94.55 | Improved |
| increase feature bank size K per identity from 1 to 5 (ablation) | ICD-Face · IJB-C (TAR@FAR=1e-4) | TAR (%): 92.73 -> 93.57 | Improved |
| increase K beyond 5 to 7 (ablation) | ICD-Face · IJB-C (TAR@FAR=1e-4) | TAR (%): 93.57 -> 93.59 | No clear effect |
| increase validity window U from 50 to 200 (ablation) | ICD-Face · IJB-C (TAR@FAR=1e-4) | TAR (%): 93.23 -> 93.57 | Improved |
| increase validity window U beyond 200 to 500 (ablation) | ICD-Face · IJB-C (TAR@FAR=1e-4) | TAR (%): 93.57 -> 93.54 | Hurt |
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.
- CoupleFace — CoupleFace: Relation Matters for Face Recognition Distillation (ECCV 2022) · better on 6 of 6 shared results
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
- Rethinking Feature-Based Knowledge Distillation for Face Recognition (CVPR 2023 · similarity 0.89)
- CoupleFace: Relation Matters for Face Recognition Distillation (ECCV 2022 · similarity 0.88)
- Evaluation-Oriented Knowledge Distillation for Deep Face Recognition (CVPR 2022 · similarity 0.88)
- Debiased Distillation for Consistency Regularization (AAAI 2025 · similarity 0.86)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.86)
- Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners (ICLR 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.
