GPU Boosted Deep Learning in Real-time Face Alignment

Peiyi Li, Xiaowei Liao, Yong Xu, Haibin Ling, Chunyuan Liao

Abstract: Research in Deep Learning (DL) has made great progresses in FA. Implementation of Convolutional Neural Networks (CNN) has fulfilled this demand, and thus made it available to some applications.• Training a CNN requires massive computing time, which is hardly achievable without GPU technology. Specifically, CNN training can run in parallel on neurons, and the training procedure is identical and independent on different neurons. The procedure is thus natural and perfect for GPU technology.