This paper addresses the challenge of collecting large-scale face datasets for improving face recognition performance. The authors propose a semi-automatic method to collect face images from the Internet and build a dataset called CASIA-WebFace, which contains approximately 10,000 subjects and 500,000 images. This dataset is designed to address the lack of publicly available large-scale face datasets, which has hindered the development of face recognition algorithms. The paper also introduces an 11-layer deep convolutional neural network (CNN) to learn discriminative representations from the CASIA-WebFace dataset. The network achieves state-of-the-art accuracy on the LFW and YouTube Faces (YTF) datasets, outperforming existing methods such as DeepFace and DeepID2. The publication of CASIA-WebFace is expected to attract more researchers to the field and accelerate the development of face recognition technology. The paper outlines the dataset construction process, the CNN architecture, and experimental results, demonstrating the effectiveness of the proposed approach.This paper addresses the challenge of collecting large-scale face datasets for improving face recognition performance. The authors propose a semi-automatic method to collect face images from the Internet and build a dataset called CASIA-WebFace, which contains approximately 10,000 subjects and 500,000 images. This dataset is designed to address the lack of publicly available large-scale face datasets, which has hindered the development of face recognition algorithms. The paper also introduces an 11-layer deep convolutional neural network (CNN) to learn discriminative representations from the CASIA-WebFace dataset. The network achieves state-of-the-art accuracy on the LFW and YouTube Faces (YTF) datasets, outperforming existing methods such as DeepFace and DeepID2. The publication of CASIA-WebFace is expected to attract more researchers to the field and accelerate the development of face recognition technology. The paper outlines the dataset construction process, the CNN architecture, and experimental results, demonstrating the effectiveness of the proposed approach.