Abstract
Facial landmark localization plays a critical role in face recognition andanalysis. In this paper, we propose a novel cascaded backbone-branches fullyconvolutional neural network~(BB-FCN) for rapidly and accurately localizingfacial landmarks in unconstrained and cluttered settings. Our proposed BB-FCNgenerates facial landmark response maps directly from raw images without anypreprocessing. BB-FCN follows a coarse-to-fine cascaded pipeline, whichconsists of a backbone network for roughly detecting the locations of allfacial landmarks and one branch network for each type of detected landmark forfurther refining their locations. Furthermore, to facilitate the faciallandmark localization under unconstrained settings, we propose a large-scalebenchmark named SYSU16K, which contains 16000 faces with large variations inpose, expression, illumination and resolution. Extensive experimentalevaluations demonstrate that our proposed BB-FCN can significantly outperformthe state-of-the-art under both constrained (i.e., within detected facialregions only) and unconstrained settings. We further confirm that high-qualityfacial landmarks localized with our proposed network can also improve theprecision and recall of face detection.