Abstract
Deep Neural Networks (DNN) have contributed a significant performance improvement in face detection. However, since most models focus only on the improvement of detection accuracy with computationally expensive structures, it is still far from real-time applications with a fast face detector. The goal of this paper is to improve face detection performance from the speed-focusing point of view. To this end, we propose a novel Fast and Accurate Face Detector (FAFD) to achieve high performance on both speed and accuracy performance. Specifically, based on the YOLOv5 model, we add one prediction head to increase the detection performance, especially for small faces. In addition, to increase the detection performance of multi-scale faces, we propose to add a novel Multi-Scale Image Fusion (MSIF) layer to the backbone network. We also propose an improved Copy-Paste to augment the training images with face objects in various scales. Experimental results on the WiderFace dataset show that the proposed FAFD achieves the best performance among the existing methods in a Speed-Focusing group. On three sub-datasets of WiderFace (i.e., Easy, Medium, and Hard sub-datasets), our FAFD yields average precisions (AP) of 95.0%, 93.5%, and 87.0%, respectively. Also, the speed performance of the FAFD is fast enough to be included in the group of speed-focusing methods.
Highlights
Face detection is an important first step in face image analysis problems such as face alignment [1,2], face recognition [3,4], and face attribute analysis [5,6]
Experimental results show that the proposed Fast and Accurate Face Detector (FAFD) achieved the best results in terms of accuracy–speed performance among the speed-focusing methods
For a high-speed face detector with CPU, we have proposed a Fast and Accurate Face Detector (FAFD) based on YOLOv5
Summary
Face detection is an important first step in face image analysis problems such as face alignment [1,2], face recognition [3,4], and face attribute analysis [5,6]. In [8–10], they reported performance improvement in a multi-task manner by combining the facial landmark localization with the existing object detector Another group of methods mainly focused on improving the architecture of CNN-based networks [11–16]. Since there are increasing demands on face detection applications in mobile environments with low computational resources equipped with CPU and edge-computing devices [19], it is important to maintain a trade-off between accuracy and speed To this end, based on the YOLOv5 object detector [20], we propose a novel Fast and Accurate Face Detector (FAFD). Our FAFD-Nano model achieved 93.0%, 91.1%, and 84.1% APs in Easy, Medium, and Hard sub-datasets, respectively These results are the best ones among the speed-focusing methods
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
