Racial Faces in the Wild: Reducing Racial Bias by Information Maximization Adaptation Network

By Mei Wang, Weihong Deng, Jiani Hu, Xunqiang Tao, Yaohai Huang. IEEE International Conference on Computer Vision, 2018

390 citations recorded by the source corpus.

Racial bias is an important issue in biometric, but has not been thoroughly studied in deep face recognition. In this paper, we first contribute a dedicated dataset called Racial Faces in-the-Wild (RFW) database, on which we firmly validated the racial bias of four commercial APIs and four state-of-the-art (SOTA) algorithms. Then, we further present the solution using deep unsupervised domain adaptation and propose a deep information maximization adaptation network (IMAN) to alleviate this bias by using Caucasian as source domain and other races as target domains. This unsupervised method simultaneously aligns global distribution to decrease race gap at domain-level, and learns the discriminative target representations at cluster level. A novel mutual information loss is proposed to further enhance the discriminative ability of network output without label information. Extensive experiments on RFW, GBU, and IJB-A databases show that IMAN successfully learns features that generalize well across different races and across different databases.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/59c47e49d8211953b1acd68984650b807ce69a71

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Cross-Domain Similarity Learning for Face Recognition in Unseen Domains (2021, methodology): “For evaluation, there exist two recent datasets recommended for studying facial cross ethnicity recognition performances, namely the Cross-Ethnicity Faces (CEF) [39] and Racial Faces in-the-Wild (RFW) [44] datasets.”
Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (2019, background): “Classical literature on mitigating bias generally operates on simpler (often linear) models [11, 50, 28], which are easier to understand and control; only recently have researchers begun looking at mitigating bias in end-to-end trained deep learning models [16, 2, 40, 18, 48, 25, 30, 36, 47, 17].”
EMOCA: Emotion Driven Monocular Face Capture and Animation (2022, methodology): “Large image datasets used by previous 3D reconstruction methods, while containing a large number of subjects of diverse ethnicities, lack emotional expressivity [14, 17, 44, 92].”
Reduced, Reused and Recycled: The Life of a Dataset in Machine Learning Research (2021, background): “Here, we see a significant amount of high stakes research being concentrated 321 on a small number of datasets, many of which contain significant racial and gender biases [4, 43].”
One Label, One Billion Faces: Usage and Consistency of Racial Categories in Computer Vision (2021, methodology): “We use four fair computer vision datasets: FairFace [28], BFW [41], RFW [49], and LAOFIW [2].”
Deep Face Recognition: A Survey (2018, methodology): “IMAN [173] simultaneously aligned global distribution to decrease race gap at domain-level, and learned the discriminative target representations at cluster level.”
WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition (2021, methodology): “Popular evaluations for face recognition including LFW families [26, 88, 63], CFP [49], AgeDB [37], RFW [70] MegaFace [29], IJB families [30, 74, 36] mainly target the pursuit of the accuracy, which have been almost saturated recently.”
Learning an animatable detailed 3D face model from in-the-wild images (2020, methodology): “BUPT-Balancedface [Wang et al. 2019] offers 7 𝑘 subjects per ethnicity (i.e. Caucasian, Indian, Asian and African), and VoxCeleb2 [Chung et al. 2018a] contains 145 𝑘 videos of 6 𝑘 subjects.”
Quantifying Societal Bias Amplification in Image Captioning (2022, neutral): “With an increased interest in fairness, multiple methods for mitigating the effects of dataset bias have been proposed [19, 32, 39, 40, 47].”