Research map / Label-efficient and robust learning / Recognition and long-tail
Learning to Attract and Repel: Dual Quality Margin Learning for Face Recognition (DQM-Face)
El Ouanas Belabbaci, Bhavesh Wani, Philipp Terhörst
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 semantic quality to magnitude quality (alpha 0 -> 0.5) (ablation) | DQM-Face · Single-image benchmarks avg | accuracy (%): 97.38 -> 97.58 | Improved |
| fuse magnitude and semantic quality instead of semantic only (ablation) | DQM-Face · Single-image benchmarks avg | accuracy (%): 97.48 -> 97.58 | Improved |
| add repulsion margin m2 with staged schedule (ablation) | DQM-Face · Single-image benchmarks avg | accuracy (%): 97.32 -> 97.58 | Improved |
| add repulsion margin m2 with staged schedule (ablation) | DQM-Face · IJB-B TAR@FAR=1e-5 | TAR (%): 87.21 -> 90.68 | Improved |
| use fixed strong repulsion margin 0.20 (ablation) | DQM-Face · Single-image benchmarks avg | accuracy (%): 97.58 -> 91.49 | Hurt |
| staged m2 schedule instead of linear schedule (ablation) | DQM-Face · Single-image benchmarks avg | accuracy (%): 97.38 -> 97.58 | Improved |
| staged m2 schedule instead of cosine schedule (ablation) | DQM-Face · IJB-B TAR@FAR=1e-5 | TAR (%): 89.20 -> 90.68 | Improved |
| staged m2 schedule instead of linear schedule (ablation) | DQM-Face · IJB-B TAR@FAR=1e-5 | TAR (%): 90.97 -> 90.68 | Hurt |
| use DQM-Face instead of MagFace on IJB-B | MagFace · IJB-B TAR@FAR=1e-5 | TAR (%): 90.36 -> 91.35 | Improved |
| use DQM-Face quality for FIQA instead of MagFace magnitude | MagFace · FIQA LFW | pAUC at 30% discard: 0.0076 -> 0.0061 | Improved |
| use DQM-Face instead of ElasticFace on single-image benchmarks | ElasticFace-Arc+ · Single-image benchmarks avg | accuracy (%): 97.53 -> 97.58 | No clear effect |
| use semantic-only quality for FIQA (ablation) | DQM-Face · FIQA XQLFW | pAUC at 30% discard: 0.1373 -> 0.1346 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- MagFace — MagFace: A Universal Representation for Face Recognition and Quality Assessment (CVPR 2021) · better on 6 of 6 shared results
- AdaFace — AdaFace: Quality Adaptive Margin for Face Recognition (CVPR 2022) · better on 1 of 1 shared result
- GroupFace — GroupFace: Learning Latent Groups and Constructing Group-Based Representations for Face Recognition (CVPR 2020) · better on 1 of 1 shared result
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · 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
- AdaFace: Quality Adaptive Margin for Face Recognition (CVPR 2022 · similarity 0.90)
- AdaptiveFace: Adaptive Margin and Sampling for Face Recognition (CVPR 2019 · similarity 0.89)
- Sub-center ArcFace: Boosting Face Recognition by Large-scale Noisy Web Faces (ECCV 2020 · similarity 0.87)
- SER-FIQ: Unsupervised Estimation of Face Image Quality Based on Stochastic Embedding Robustness (CVPR 2020 · similarity 0.87)
- Attributes Shape the Embedding Space of Face Recognition Models (ICML 2025 · similarity 0.85)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · 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.
