Research map / Label-efficient and robust learning / Domain adaptation and generalization
Racial Faces in the Wild: Reducing Racial Bias by Information Maximization Adaptation Network
Mei Wang, Weihong Deng, Jiani Hu, Xunqiang Tao, Yaohai Huang
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 clustering-based pseudo-label pre-adaptation (ablation) | IMAN-A · RFW African target (6000 pairs) | verification accuracy (%): 85.52 -> 88.12 | Improved |
| add MI loss on top of pseudo-adaptation (ablation) | IMAN-A · RFW Indian target (6000 pairs) | verification accuracy (%): 92.08 -> 93.55 | Improved |
| add MI loss on top of pseudo-adaptation (ablation) | IMAN-A · RFW African target (6000 pairs) | verification accuracy (%): 88.12 -> 88.88 | Improved |
| add MI loss on top of pseudo-adaptation (ablation) | IMAN-A · RFW Asian target (6000 pairs) | verification accuracy (%): 88.80 -> 89.87 | Improved |
| add pseudo-label adaptation to MI-only adaptation (MI alone with random classifier) (ablation) | IMAN-A · RFW Asian target (6000 pairs) | verification accuracy (%): 86.88 -> 89.87 | Improved |
| use IMAN-A instead of Arcface baseline | Arcface (CASIA source) · RFW African target (6000 pairs) | verification accuracy (%): 85.13 -> 88.88 | Improved |
| replace Softmax with Arcface loss in pre-adaptation stage (IMAN*-A) | IMAN-A · RFW African target (6000 pairs) | verification accuracy (%): 88.88 -> 91.42 | Improved |
| use MMD-based DAN-A instead of plain Arcface | Arcface · RFW African target (6000 pairs) | verification accuracy (%): 85.13 -> 86.30 | Improved |
| train on race-balanced Equalizedface instead of CASIA-WebFace | Arcface ResNet-34 · RFW African (6000 pairs) | verification accuracy (%): 84.93 -> 90.98 | Improved |
| train on race-balanced Equalizedface instead of CASIA-WebFace | Arcface ResNet-34 · CFP-FP | accuracy (%): 93.91 -> 92.74 | Hurt |
| train on single-race 7000 identities instead of balanced 3.5K per race (Asian) | Arcface ResNet-34 · RFW Asian (6000 pairs) | verification accuracy (%): 90.60 -> 91.27 | Mixed |
| apply IMAN-A adaptation (CASIA to GBU) instead of Arcface | Arcface(CASIA) · GBU Ugly | VR at FAR 0.001 (%): 75.00 -> 85.38 | Improved |
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 3 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
- Mitigating Bias in Face Recognition Using Skewness-Aware Reinforcement Learning (CVPR 2020 · similarity 0.89)
- Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation (ICCV 2023 · similarity 0.87)
- Mitigating Gender Bias in Face Recognition using the von Mises-Fisher Mixture Model (ICML 2022 · similarity 0.86)
- Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face Recognition (NeurIPS 2023 · similarity 0.86)
- Meta Balanced Network for Fair Face Recognition (TPAMI 2021 · similarity 0.86)
- Towards Universal Representation Learning for Deep Face Recognition (CVPR 2020 · 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.
