Research map / Label-efficient and robust learning / Recognition and long-tail
Meta Balanced Network for Fair Face Recognition
Mei Wang, Yaobin Zhang, Weihong Deng
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 MBN(arc) instead of Arcface | MBN · BUPT-Globalface-4 / IDS-4 | SER: 2.33 -> 1.59 | Improved |
| use MBN(cos) instead of Cosface | MBN · BUPT-Globalface-8 / IDS-8 | SER: 2.28 -> 1.55 | Improved |
| use MBN(arc) instead of Arcface | MBN · BUPT-Balancedface-4 / IDS-4 | SER: 1.65 -> 1.37 | Improved |
| learn adaptive margins per skin-tone group (MBN(arc)) instead of uniform margin | MBN · BUPT-Globalface-4 / IDS-4 | average accuracy: 95.37 -> 95.93 | Improved |
| MBN(soft) vs Norm-Softmax on skewed training distribution 7:1:1:1 | MBN · simulated Globalface-4 | STD: 3.46 -> 2.19 | Improved |
| MBN(soft) vs Norm-Softmax on training distribution 4:2:2:2 | MBN · simulated Globalface-4 | STD: 2.64 -> 1.33 | Improved |
| train on skin-tone-balanced data instead of Globalface | Arcface · IDS-8 | STD: 1.52 -> 1.16 | Improved |
| train on skin-tone-balanced Balancedface* instead of CASIA-WebFace | Arcface · IDS-4 Tone-IV | accuracy: 84.93 -> 90.98 | Improved |
| use sample re-weighting instead of MBN | Arcface · BUPT-Globalface-4 / IDS-4 | SER: 1.79 -> 1.59 | Hurt |
| use adversarial attribute removal instead of MBN | Arcface · BUPT-Globalface-4 / IDS-4 | SER: 1.87 -> 1.59 | Hurt |
| use MBN instead of RL-RBN for margin learning | MBN(arc) · BUPT-Globalface-4 / IDS-4 | SER: 1.76 -> 1.59 | Hurt |
| use MBN instead of Arcface on age cohorts of Tone-VI | MBN · IDS-4 Tone-VI age cohorts | STD: 3.21 -> 0.94 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Arcface — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 9 of 9 shared results
- Cosface — CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018) · 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 0.88)
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.86)
- Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face Recognition (NeurIPS 2023 · similarity 0.86)
- FairCal: Fairness Calibration for Face Verification (ICLR 2022 · similarity 0.86)
- Racial Faces in the Wild: Reducing Racial Bias by Information Maximization Adaptation Network (ICCV 2019 · similarity 0.86)
- Learning To Learn Across Diverse Data Biases in Deep Face Recognition (CVPR 2022 · 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.
