Fast Face-CPU: A Real-time Fast Face Detector on CPU Using Deep Learning
Muhamad Dwisnanto Putro, Kang-Hyun Jo · 2020
Face detection is a vision method that can detect the human faces in an image. This method plays a fundamental role in the success of human face classifications. Besides, the practical applications require to work in real-time on low-cost devices. Several traditional methods have implemented it, but there are constraints on accuracy. On the other hand, the success of deep learning method for extracting object features that can also be applied to distinguish faces and backgrounds. However, Deep Convolutional Neural Network (DCNN) requires heavy computation and tends to be slow when implemented on the CPU. The work in this paper builds the Fast Face-CPU (FFCPU) is a real-time face detector that can work on the CPU. The proposed architecture using the light CNN, which consists of two crucial modules, including the backbone to quickly extract facial features and the detection module as a predictor of multiple faces. As a result, the FFCPU obtains a superior speed by 209 frames per second on CPU and achieves state-of-the-art performances on several benchmark datasets when compared with other CPU-based models.