The paper "DeepID3: Face Recognition with Very Deep Neural Networks" by Yi Sun, Ding Liang, Xiaogang Wang, and Xiaouu Tang introduces two very deep neural network architectures, DeepID3, for face recognition. These architectures are derived from the VGG net and GoogLeNet, incorporating stacked convolution and inception layers to enhance their suitability for face recognition tasks. The networks are trained using joint face identification-verification supervisory signals, which are added to both intermediate and final feature extraction layers. The proposed architectures achieve a face verification accuracy of 99.53% and a rank-1 face identification accuracy of 96.0% on the LFW dataset, surpassing the performance of the previous state-of-the-art, DeepID2+. The paper also discusses the limitations of DeepID3, particularly in handling wrong labels, and suggests that larger-scale training data could further improve the performance of very deep neural networks in face recognition.The paper "DeepID3: Face Recognition with Very Deep Neural Networks" by Yi Sun, Ding Liang, Xiaogang Wang, and Xiaouu Tang introduces two very deep neural network architectures, DeepID3, for face recognition. These architectures are derived from the VGG net and GoogLeNet, incorporating stacked convolution and inception layers to enhance their suitability for face recognition tasks. The networks are trained using joint face identification-verification supervisory signals, which are added to both intermediate and final feature extraction layers. The proposed architectures achieve a face verification accuracy of 99.53% and a rank-1 face identification accuracy of 96.0% on the LFW dataset, surpassing the performance of the previous state-of-the-art, DeepID2+. The paper also discusses the limitations of DeepID3, particularly in handling wrong labels, and suggests that larger-scale training data could further improve the performance of very deep neural networks in face recognition.