Research map / Label-efficient and robust learning / Recognition and long-tail
Fair Loss: Margin-Aware Reinforcement Learning for Deep Face Recognition
Bingyu Liu, Weihong Deng, Yaoyao Zhong, Mei Wang, 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 learned adaptive margin (Fair Loss) to CosFace | CosFace, 34-layer · MegaFace Challenge 1 verification | TAR (%): 89.15 -> 90.82 | Improved |
| add learned adaptive margin (Fair Loss) to ArcFace | ArcFace, 34-layer · MegaFace Challenge 1 verification | TAR (%): 91.06 -> 92.89 | Improved |
| add learned adaptive margin (Fair Loss) to CosFace | CosFace, 18-layer · MegaFace Challenge 1 verification | TAR (%): 86.80 -> 89.14 | Improved |
| add learned adaptive margin (Fair Loss) to ArcFace | ArcFace, 18-layer · YTF | accuracy (%): 93.36 -> 93.76 | Improved |
| add learned adaptive margin (Fair Loss) to CosFace | CosFace, 34-layer · LFW | accuracy (%): 99.32 -> 99.52 | Improved |
| use RL-learned margin strategy instead of margins inversely proportional to class size (MIP) (ablation) | CosFace · MegaFace Challenge 1 verification | TAR (%): 91.21 -> 92.87 | Improved |
| use RL-learned margin strategy instead of variance-deciding margins (VDM) (ablation) | CosFace · MegaFace Challenge 1 verification | TAR (%): 91.43 -> 92.87 | Improved |
| use RL-learned margin strategy instead of random margins (RM) (ablation) | CosFace · MegaFace Challenge 1 verification | TAR (%): 91.36 -> 92.87 | Improved |
| use RL-learned margin strategy instead of rule-based or random margins (ablation) | CosFace · MegaFace Challenge 1 rank-1 | rank-1 accuracy (%): 76.77 -> 77.45 | Mixed |
| use RL-learned margin strategy instead of rule-based margins (ablation) | CosFace · YTF | accuracy (%): 95.4 -> 96.2 | Mixed |
| add Fair Loss to CosFace under the large MF2 protocol | CosFace · MegaFace Challenge 2 rank-1 | rank-1 accuracy (%): 74.11 -> 79.41 | Improved |
| add Fair Loss to CosFace under the large MF2 protocol | CosFace · MegaFace Challenge 2 verification | TAR (%): 86.77 -> 89.62 | 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 4 of 4 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
- Loss Function Search for Face Recognition (ICML 2020 · similarity 0.90)
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.90)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.89)
- AdaptiveFace: Adaptive Margin and Sampling for Face Recognition (CVPR 2019 · similarity 0.88)
- Learning Towards The Largest Margins (ICLR 2022 · similarity 0.88)
- CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition (CVPR 2020 · similarity 0.88)
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.
