Okcu Sefa Burak ; Özkalaycı Burak Oğuz ; Çığla Cevahir

Abstract

Face recognition applications in practice are composed of two main steps:face detection and feature extraction. In a sole vision-based solution, thefirst step generates multiple detection for a single identity by ingesting acamera stream. A practical approach on edge devices should prioritize thesedetection of identities according to their conformity to recognition. In thisperspective, we propose a face quality score regression by just appending asingle layer to a face landmark detection network. With almost no additionalcost, face quality scores are obtained by training this single layer to regressrecognition scores with surveillance like augmentations. We implemented theproposed approach on edge GPUs with all face detection pipeline steps,including detection, tracking, and alignment. Comprehensive experiments showthe proposed approach's efficiency through comparison with SOTA face qualityregression models on different data sets and real-life scenarios.