Research map / Label-efficient and robust learning / Recognition and long-tail
FairCal: Fairness Calibration for Face Verification
Tiago Salvador, Stephanie Cairns, Vikram Voleti, Noah Marshall, Adam M Oberman
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 |
|---|---|---|---|
| Use FairCal’s feature-cluster-conditional calibration instead of cosine-score-only calibration | FaceNet face verification · RFW | AUROC: 88.26 -> 90.58 | Improved |
| Use FairCal’s feature-cluster-conditional calibration instead of cosine-score-only calibration | FaceNet face verification · RFW | mean KS: 6.37 -> 1.37 | Improved |
| Use FairCal’s feature-cluster-conditional calibration instead of cosine-score-only calibration | FaceNet face verification · BFW | AUROC: 96.06 -> 96.90 | Improved |
| Use FairCal’s feature-cluster-conditional calibration instead of cosine-score-only calibration | FaceNet face verification · BFW | mean KS: 6.77 -> 3.09 | Hurt |
| Use FairCal rather than the baseline | ArcFace face verification · BFW | AUROC: 97.41 -> 97.44 | Improved |
| Use FairCal rather than the baseline | ArcFace face verification · BFW | mean KS: 2.57 -> 2.49 | Improved |
| Use FairCal rather than FSN | FaceNet face verification · RFW, 1% global FPR | AAD of subgroup FPR: 0.37 -> 0.28 | Improved |
| Use FairCal rather than FSN | FaceNet face verification · RFW, 1% global FPR | AAD of subgroup FPR: 0.35 -> 0.29 | Improved |
| Use FairCal rather than the strongest listed baseline | Face verification · BFW, 1% global FPR, FaceNet (Webface) | AAD of subgroup FPR: 0.77 -> 0.80 | Hurt |
| Use FairCal rather than the strongest listed baseline | ArcFace face verification · BFW, 1% global FPR | AAD of subgroup FPR: 0.44 -> 0.63 | Hurt |
| Use FairCal rather than the baseline | Face verification · IJB-C, FaceNet (VGGFace2) | AUROC: 92.72 -> 94.74 | Improved |
| Use FairCal rather than the baseline | Face verification · IJB-C, FaceNet (Webface) | AUROC: 89.23 -> 92.04 | Improved |
Closest papers by meaning
- [Re] On the Reproducibility of “FairCal: Fairness Calibration for Face Verification” (NeurIPS 2023 · similarity 0.89)
- Assessing Uncertainty in Similarity Scoring: Performance & Fairness in Face Recognition (ICLR 2024 · similarity 0.86)
- Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning (NeurIPS 2025 · similarity 0.85)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (AAAI 2024 · similarity 0.85)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.85)
- FairDD: Fair Dataset Distillation (NeurIPS 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.
