A Novel BlazeFace Based Pre-processing for MobileFaceNet in Face Verification
Necmettin Bayar, Kubra Guzel, Deniz Kumlu · 2022
Face verification is an important security step on mobile devices and many other systems, thus it has to work with high accuracy. Besides the importance of the accuracy in the face verification model, its weight and computational complexity play crucial roles especially in mobile devices. In this study, we aimed to provide novel contribution as pre-processing step for MobileFaceNet without affecting its accuracy. With this contri-bution, overall pipeline has smaller weight and faster inference time by comparison to the available pre-processing models for MobileFaceNet such as multi task cascaded convolutional neural network (MTCNN) and RetinaFace. The face verification test results show the superiority of our proposed model compared to the state-of-the-art models in terms of weight and speed.