Research map / Label-efficient and robust learning / Recognition and long-tail
Mitigating Face Recognition Bias via Group Adaptive Classifier
Sixue Gong, Xiaoming Liu, Anil K. Jain
Check 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 |
|---|---|---|---|
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · RFW | STD across groups: 0.96 -> 0.58 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · RFW | Avg verification accuracy (%): 94.64 -> 95.21 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · LFW | Verification accuracy (%): 99.75 -> 99.78 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · IJB-A (0.1% FAR) | Verification accuracy (%): 90.2 ± 1.1 -> 91.3 ± 1.2 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · IJB-C (0.001% FAR) | Verification accuracy (%): 80.2 -> 83.5 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · IJB-C (0.01% FAR) | Verification accuracy (%): 88.0 -> 89.2 | Improved |
| Use group-adaptive kernels and attention with automatic layer selection | ArcFace-50 · IJB-C (0.1% FAR) | Verification accuracy (%): 92.9 -> 93.7 | Improved |
| Use channel-wise attention only (ablation) | GAC · RFW | STD across groups: 1.11 -> 0.83 | Improved |
| Use adaptive kernels only (ablation) | GAC · RFW | STD across groups: 1.11 -> 0.78 | Improved |
| Use spatial attention only (ablation) | GAC · RFW | Avg verification accuracy (%): 94.64 -> 94.19 | Hurt |
| Add spatial attention to the adaptive kernels and channel-wise attention (ablation) | GAC · RFW | Avg verification accuracy (%): 95.21 -> 94.72 | Hurt |
| Set the adaptation threshold to τ=-0.1 instead of τ=-0.2 (ablation) | GAC · RFW | Avg verification accuracy (%): 95.21 -> 95.01 | 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.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 2 of 3 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
- Jointly De-biasing Face Recognition and Demographic Attribute Estimation (ECCV 2020 · similarity 0.89)
- Consistent Instance False Positive Improves Fairness in Face Recognition (CVPR 2021 · similarity 0.89)
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.89)
- Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face Recognition (NeurIPS 2023 · similarity 0.88)
- Mitigating Gender Bias in Face Recognition using the von Mises-Fisher Mixture Model (ICML 2022 · similarity 0.87)
- GroupFace: Learning Latent Groups and Constructing Group-Based Representations for Face Recognition (CVPR 2020 · similarity 0.86)
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.
