Abstract
With the rapid development of smartphones, facial analysis has been playingan increasingly important role in a multitude of mobile applications. In mostscenarios, face tracking serves as a crucial first step because more often thannot, a mobile application would only need to focus on analysing a specific facein a complex setting. Albeit inheriting many commons traits of the genericvisual tracking problem, face tracking in mobile scenarios is characterised bya unique set of challenges. In this work, we propose iBUG MobiFace benchmark,the first mobile face tracking benchmark consisting of 50 sequences captured bysmartphone users in unconstrained environments. The sequences contain a totalof 50,736 frames with 46 distinct identities to be tracked. The tracking targetin each sequence is selected with varying difficulties in mobile scenarios. Inaddition to frame by frame bounding box, the annotations of 9 sequenceattributes(e.g. multiple faces) are provided. We further provide a survey of 23state-of-the-art visual trackers and a comprehensive quantitative evaluation ofthese methods on the proposed benchmark. In particular, trackers from two mostpopular frameworks, namely, correlation filter-based tracking and deeplearning-based tracking, are studied. Our experiment shows that (a) theperformance of all existing generic object trackers drops significantly on themobile face tracking scenario, suggesting the need of more research effort intomobile face tracking, and (b) the effective combination of deep learningtracking and face-related algorithms(e.g. face detection) provides the mostpromising basis for future developments in the field. The database, annotationsand evaluation protocol/code will be made publicly available on the iBUGwebsite.