Research map / Label-efficient and robust learning / Recognition and long-tail
Jointly De-biasing Face Recognition and Demographic Attribute Estimation
Sixue Gong, Xiaoming Liu, Anil K. Jain
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 adversarial disentanglement (DebFace) to the baseline face encoder | face verification · Verification biasness - All | biasness: 6.83 -> 5.07 | Improved |
| add adversarial disentanglement (DebFace) to the baseline | face verification · Verification biasness - Gender | biasness: 0.50 -> 0.15 | Improved |
| add adversarial disentanglement (DebFace) to the baseline | face verification · Verification biasness - Age | biasness: 3.13 -> 1.83 | Improved |
| add adversarial disentanglement (DebFace) to the baseline | face verification · Verification biasness - Race | biasness: 5.49 -> 3.70 | Improved |
| add joint adversarial de-biasing to the gender estimator | demographic estimation · Gender estimation biasness | biasness: 12.38 -> 10.22 | Improved |
| add joint adversarial de-biasing to the age estimator | demographic estimation · Age estimation biasness | biasness: 10.83 -> 7.61 | Improved |
| add joint adversarial de-biasing to the race estimator | demographic estimation · Race estimation biasness | biasness: 14.58 -> 10.00 | Improved |
| use de-biased DebFace-ID features instead of BaseFace features | face verification · LFW | accuracy (%): 99.38 -> 98.97 | Hurt |
| use de-biased DebFace-ID features instead of BaseFace features | face verification · IJB-A | TAR @ 0.1% FAR (%): 90.2 ± 1.1 -> 87.6 ± 0.9 | Hurt |
| use DebFace-ID instead of BaseFace | face verification · IJB-C | TAR @ 0.001% FAR (%): 80.2 -> 82.0 | Mixed |
| aggregate gender and race features with DebFace-ID (DemoID) instead of DebFace-ID alone | face verification · LFW | accuracy (%): 98.97 -> 99.50 | Improved |
| aggregate gender and race features with DebFace-ID (DemoID) instead of DebFace-ID alone | face verification · IJB-A | TAR @ 0.1% FAR (%): 87.6 ± 0.9 -> 92.2 ± 0.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 0 of 1 shared result
- 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
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.89)
- Mitigating Face Recognition Bias via Group Adaptive Classifier (CVPR 2021 · similarity 0.89)
- Consistent Instance False Positive Improves Fairness in Face Recognition (CVPR 2021 · similarity 0.89)
- AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (CVPR 2025 · similarity 0.85)
- Balancing Act: Distribution-Guided Debiasing in Diffusion Models (CVPR 2024 · similarity 0.85)
- MABR: Multilayer Adversarial Bias Removal Without Prior Bias Knowledge (AAAI 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.
