基于动态样本策略的深度人脸识别
首发时间:2017-11-20
摘要:近年来,深度卷积神经网络在目标分类、物体检测等任务中均获得了很突出的结果。基于深度卷积神经网络的人脸识别也在公开数据集中超过了人眼的识别能力。在人脸特征提取的训练中,特征嵌入已经成为了比较热门的方法。本文基于一个深度特征嵌入的卷积神经网络,提出了一种样本选取策略,得到了更有区分能力的特征。通过简单的人脸对齐和模型融合,在LFW和MegeFace数据集上,达到了不错的识别结果。
关键词: 人工智能 深度卷积神经网络 人脸识别 动态样本策略
For information in English, please click here
Deep Face Recognition via Dynamic Example Strategy
Abstract:Recently, deep convolutional neutral network has made great progress in object classification and object detection。And face recognition system based on deep CNNs has already exceed human’s performance in many public datasets。Feature embedding method is popular in face representation learning。This paper proposes a dynamic example strategy based on deep embedding CNNs to learn more discriminative face representations. At last, a state-of-the-art performance is achieved on LFW and MegaFace with only weekly aligned faces.
引用
No.****
同行评议
共计0人参与
勘误表
基于动态样本策略的深度人脸识别
评论
全部评论0/1000