Research map / Label-efficient and robust learning / Recognition and long-tail
Invariant Feature Regularization for Fair Face Recognition
Jiali Ma, Zhongqi Yue, Kagaya Tomoyuki, Suzuki Tomoki, Karlekar Jayashree, Sugiri Pranata, Hanwang Zhang
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 Inv-Reg to Arcface | Arcface ResNet-100 · MFR | average accuracy (%): 76.90 -> 79.44 | Improved |
| add Inv-Reg to CIFP | CIFP ResNet-100 · MFR | average accuracy (%): 80.32 -> 81.17 | Improved |
| add Inv-Reg to Arcface | Arcface ResNet-50 · MFR | African accuracy (%): 74.54 -> 77.00 | Improved |
| add Inv-Reg to Arcface | Arcface ResNet-100 · DigiFace-1M | verification accuracy FPR=1e-4 (%): 64.32 -> 67.57 | Improved |
| add Inv-Reg to Arcface | Arcface ResNet-100 · IJB-C | verification accuracy (%): 95.60 -> 96.46 | Improved |
| use learned partitions instead of ground-truth race and gender partitions (ablation) | Inv-Reg · MFR (ResNet-50) | average accuracy (%): 74.84 -> 75.03 | Improved |
| use ground-truth partitions instead of no invariance (Arcface baseline) (ablation) | Inv-Reg · MFR (ResNet-50) | average accuracy (%): 73.43 -> 74.84 | Improved |
| increase number of subsets K from 2 to 4 (ablation) | Inv-Reg · MFR (ResNet-50) | average accuracy (%): 77.19 -> 76.94 | Hurt |
| increase discovered partitions from 1 to 3 (ablation) | Inv-Reg · MFR (ResNet-50) | average accuracy (%): 76.75 -> 77.19 | Improved |
| increase discovered partitions from 3 to 4 (ablation) | Inv-Reg · MFR (ResNet-50) | average accuracy (%): 77.19 -> 76.82 | Hurt |
| vary invariant penalty weight lambda between 0.02 and 0.2 (ablation) | Inv-Reg · MFR (ResNet-50) | accuracy | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Arcface — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 5 of 5 shared results
- Anchorface — AnchorFace: An Anchor-based Facial Landmark Detector Across Large Poses (AAAI 2021) · better on 1 of 2 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
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.87)
- Consistent Instance False Positive Improves Fairness in Face Recognition (CVPR 2021 · similarity 0.87)
- FairDD: Fair Dataset Distillation (NeurIPS 2025 · similarity 0.86)
- Distributionally Generative Augmentation for Fair Facial Attribute Classification (CVPR 2024 · similarity 0.85)
- Fair Deepfake Detectors Can Generalize (NeurIPS 2025 · similarity 0.84)
- AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (CVPR 2025 · similarity 0.84)
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.
