This repo contains the code accompanying the paper, Understanding and Mitigating Annotation Bias in Facial Expression Recognition (Chen and Joo, ICCV 2021).

• annotation_bias_evaluation folder contains the code that calculates the annotation bias across gender, age, and race and plots the figures in the paper and the appendix.
• auc_fer folder contains the code for AUC-FER and various baseline methods shown in Table 4 of the paper.

• FairFace: https://github.com/dchen236/FairFace
• Expression in-the-Wild (ExpW): https://mmlab.ie.cuhk.edu.hk/projects/socialrelation/
• EmotioNet: https://cbcsl.ece.ohio-state.edu/enc-2020/index.html
• Real-world Affective Faces Database (RAF-DB): http://www.whdeng.cn/raf/model1.html
• AffectNet: http://mohammadmahoor.com/affectnet/
• Karolinska Directed Emotional Faces (KDEF): https://www.kdef.se/
• Chicago Face Database (CFD): https://www.chicagofaces.org/

OpenFace: https://github.com/TadasBaltrusaitis/OpenFace

The code for AUC-FER and its baseline methods is adapted from the CVPR paper: Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation, Code