摘要
Great progress has been made toward accurate face detection in recent years.However,the heavy model and expensive computation costs make it difficult to deploy many detectors on mobile and embedded devices where model size and latency are highly constrained.In this paper,we present a millisecond-level anchor-free face detector,YuNet,which is specifically designed for edge devices.There are several key contributions in improving the efficiency-accuracy trade-off.First,we analyse the influential state-of-theart face detectors in recent years and summarize the rules to reduce the size of models.Then,a lightweight face detector,YuNet,is introduced.Our detector contains a tiny and efficient feature extraction backbone and a simplified pyramid feature fusion neck.To the best of our knowledge,YuNet has the best trade-off between accuracy and speed.It has only 75856 parameters and is less than 1/5 of other small-size detectors.In addition,a training strategy is presented for the tiny face detector,and it can effectively train models with the same distribution of the training set.The proposed YuNet achieves 81.1%mAP(single-scale)on the WIDER FACE validation hard track with a high inference efficiency(Intel i7-12700K:1.6ms per frame at 320×320).Because of its unique advantages,the repository for YuNet and its predecessors has been popular at GitHub and gained more than 11K stars at https://github.com/ShiqiYu/libfacedetection.Keywords:Face detection,object detection,computer version,lightweight,inference efficiency,anchor-free mechanism.
基金
supported in part by National Natural Science Foundation of China(No.61976144)
the Stable Support Plan Program of Shenzhen Natural Science Fund,China(No.20200925155017002)
the National Key Research and Development Program of China(No.2020 AAA0140000).
作者简介
Wei Wu received the B.Sc.degree in computer science and technology from Chongqing University,China in 2017.Currently,he is a master student in electronics science and technology at Department of Computer Science and Engineering,Southern University of Science and Technology,China.His research interests include object detection and computer vision.E-mail:12032501@mail.sustech.edu.cn ORCID iD:0000-0002-9595-1778;Hanyang Peng received the B.Sc.degree in measurement and control technology from Northeast University of China,China in 2008,the M.Eng.degree in detection technology and automatic equipment from Tianjin University,China in 2010,and the Ph.D.degree in pattern recognition and intelligence systems from Institute of Automation,Chinese Academy of Sciences,China in 2017.He currently works as an assistant professor in Pengcheng Laboratory,China.His research interests include computer vision,machine learning and distributed learning.E-mail:penghy@pcl.ac.cn ORCID iD:0000-0002-9715-473X;Shiqi Yu received the B.Eng.degree in computer science and engineering from Chu Kochen Honors College,Zhejiang University,China in 2002,and the Ph.D.degree in pattern recognition and the intelligent systems from Institute of Automation,Chinese Academy of Sciences,China in 2007.He is currently an associate professor in Department of Computer Science and Engineering,Southern University of Science and Technology,China.He worked as an assistant professor and an associate professor in Shenzhen Institutes of Advanced Technology,Chinese Academy of Sciences,China from 2007 to 2010,and as an associate professor in Shenzhen University,China from 2010 to 2019.His research interests include gait recognition,face detection and computer vision.E-mail:yusq@sustech.edu.cn(Corresponding author)ORCID iD:0000-0002-5213-5877。
