大规模亚洲人脸数据集的构建

摘要

At the use of deep learning method,the accuracy of face recognition has got breakthrough progress,and gradually become a very practical way.The promotion of it mainly relys on two aspects:optimizing the network model and enhancing the training data sets.However,the public face data sets are very scarce at present,especially for Asian face data sets.Therefore,this paper created an Asian face data set,which contains 5 thousand stars of 0.5 million images at the use of semi-automatic processing.While,it compared WebFace public data set to find that it has obvious advantages in accuracy.%随着深度学习方法的使用,人脸识别准确率得到突破性进步.基于深度学习的人脸识别方法准确率提升依赖于网络模型优化和训练数据集增强两个方面.然而,亚洲人脸的公开数据集非常稀缺,因此,文中通过利用半自动化处理流程创建了包含5k个亚洲明星(共计50w张图片)的人脸数据集,并比较了基于WebFace公开数据集与基于以上流程所建数据集训练的深度网络测试效果.相比之下,本文所建数据集在亚洲人脸的测试准确率具有明显优势.