Recognizing Disguised Faces in the Wild

Research in face recognition has seen tremendous growth over the past coupleof decades. Beginning from algorithms capable of performing recognition inconstrained environments, the current face recognition systems achieve veryhigh accuracies on large-scale unconstrained face datasets. While upcomingalgorithms continue to achieve improved performance, a majority of the facerecognition systems are susceptible to failure under disguise variations, oneof the most challenging covariate of face recognition. Most of the existingdisguise datasets contain images with limited variations, often captured incontrolled settings. This does not simulate a real world scenario, where bothintentional and unintentional unconstrained disguises are encountered by a facerecognition system. In this paper, a novel Disguised Faces in the Wild (DFW)dataset is proposed which contains over 11000 images of 1000 identities withdifferent types of disguise accessories. The dataset is collected from theInternet, resulting in unconstrained face images similar to real worldsettings. This is the first-of-a-kind dataset with the availability ofimpersonator and genuine obfuscated face images for each subject. The proposeddataset has been analyzed in terms of three levels of difficulty: (i) easy,(ii) medium, and (iii) hard in order to showcase the challenging nature of theproblem. It is our view that the research community can greatly benefit fromthe DFW dataset in terms of developing algorithms robust to such adversaries.The proposed dataset was released as part of the First International Workshopand Competition on Disguised Faces in the Wild at CVPR, 2018. This paperpresents the DFW dataset in detail, including the evaluation protocols,baseline results, performance analysis of the submissions received as part ofthe competition, and three levels of difficulties of the DFW challenge dataset.