Research map / Label-efficient and robust learning / Recognition and long-tail
Mitigating Gender Bias in Face Recognition using the von Mises-Fisher Mixture Model
Jean-Rémy Conti, Nathan Noiry, Stephan Clemencon, Vincent Despiegel, Stéphane Gentric
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 Ethical Module (EM-C, kappa0=45, kappa1=30) on top of frozen ArcFace | ArcFace ResNet50 · LFW, FAR 10-4 | BFAR and BFRR jointly: BFAR 4.72→2.44, BFRR 10.27→9.18 | Improved |
| Add Ethical Module (EM-FAR, kappa0=15, kappa1=20) | ArcFace ResNet50 · LFW, FAR 10-4 | BFAR: 4.72→2.11 | Improved |
| Add Ethical Module (EM-FRR, kappa0=25, kappa1=20) | ArcFace ResNet50 · LFW, FAR 10-4 | BFAR: 4.72→33.65 | Hurt |
| Add Ethical Module (EM-FRR) to ArcFace | ArcFace ResNet50 · LFW, FAR 10-4 | BFRR: 10.27→5.89 | Improved |
| Ethical Module (EM-FAR/EM-C) on CosFace/AdaCos/CurricularFace (MobileFaceNet) | AdaCos, CosFace, CurricularFace · LFW, FAR 10-4 | BFAR: e.g., CosFace 2.51→1.13 (EM-FAR), AdaCos 3.84→1.41 | Improved |
| Ethical Module with gender-dependent concentration (kappa per gender) trained on frozen embeddings | ArcFace ResNet50 vs PASS-G baseline · LFW, FAR 10-3 | BFAR: EM-C 1.20 vs PASS-G 2.11 vs original 1.81 | Improved |
| Ethical Module increases FRR@FAR error rate in some variants | ArcFace ResNet50 and MobileFaceNet models · LFW, FAR 10-4 | FRR@FAR: e.g., ArcFace 0.078→0.151 (EM-FAR); CosFace 1.73→3.69 (EM-FAR) | Hurt |
| Adversarial post-processing PASS-G baseline | ArcFace ResNet50 · LFW, FAR 10-4 | FRR@FAR: 0.078→0.315 | Hurt |
| Sweeping gender-specific concentration parameters (kappa0, kappa1) shows BFAR and BFRR trade off against each other (ablation) | ArcFace ResNet50 Ethical Module · IJB-C grid search, FAR 10-3 | BFAR vs BFRR | 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.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 5 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
- Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations (ICCV 2019 · similarity 0.91)
- Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face Recognition (NeurIPS 2023 · similarity 0.91)
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.90)
- Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (CVPR 2020 · similarity 0.89)
- Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation (ICCV 2023 · similarity 0.89)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (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.
