Face recognition trained on a small dataset
作者单位：Institute of Remote Test and ControlChong Qing University of technology
会议名称：《OSEC首届兵器工程大会》
会议日期：2017年
学科分类：08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程]
摘 要：Face recognition has been very successful in recent years with the rapid development of the Convolutional Neural Networks(CNNs). However, the cost of the success is the bigger datasets and the deeper networks. It is beyond the capabilities of most international research groups. In this work, we propose a method that only trains on CASIA Web Face. Our method combine the improved center loss and softmax loss as joint supervision in a network with balance the training data. The benefit of our approach is much easier and useful: we achieve comparable state of the art results on the stand Labeled Faces in the Wild(LFW) and You Tube Faces(YTF) face benchmarks with only 0.8 images.