Summary: The accuracy of state-of-the-art face verification systems has surpassed the ability of humans to recognise other humans and it is improving every year . However, datasets used for training and testing are often imbalanced and there have been numerous reports that the accuracy of trained models is significantly worse for people with gender and/or skin tones that are underrepresented in the data, hence rendering them biased [2,3]. The main objective of this workshop and its associated challenge is to advance into fairness in terms of bias mitigation in biometric/looking-at-people systems. To that end we propose an event to compile the latests efforts in the field and run a fair face verification challenge with a new large set of data and new fairness-aware evaluation metrics.
Topic and Motivation: This workshop will focus on bias analysis and mitigation methodologies, which will result into more fair face recognition and analysis systems. These advances will have a direct impact within society's equality of opportunity. In this proposal we plan to provide a comprehensive up to date review on fair face recognition and analysis research. We find of crucial interest to centralize ideas, discuss them and push the field to advance towards more fair systems for the good of society. Complementary to that, we will also contribute pushing research in the field by releasing a large annotated dataset for fair face verification and running an associated challenge. The main topics of interest related to fair face recognition and analysis are (but not limited to):
- Dealing with unbalanced and/or noisy datasets.
- Novel methodologies, metrics and algorithms for bias mitigation.
- Explainable/interpretable strategies for understanding/dealing with bias.
- Generative models applications.
Important dates and submission instructions: here.
 Surpassing human-level face verification performance on LFW with gaussianface. Chaochao Lu, Xiaoou Tang; AAAI’15 Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence Pages 3811-3819, 2015.
 Gender Shades: Intersectional Accuracy Disparities in Commercial GenderClassification. Joy Buolamwini, Timnit Gebru; Proceedings of the 1st Conferenceon Fairness, Accountability and Transparency, PMLR 81:77-91, 2018.
 Face Recognition Vendor Test (FRVT) Part 3: Demographic Effects. PatrickGrother, Mei Ngan, Kayee Hanaoka. NIST report, 2019.
 IARPA Janus Benchmark - C: Face Dataset and Protocol. Brianna Maze etal.; International Conference on Biometrics (ICB), pp. 158-165, 2018