This paper addresses the challenge of developing effective feature representations for face recognition by leveraging deep learning and using both face identification and verification signals as supervision. The authors propose DeepID2, a feature representation learned through deep convolutional networks. The identification task increases inter-personal variations by separating DeepID2 extracted from different identities, while the verification task reduces intra-personal variations by pulling DeepID2 extracted from the same identity together. This approach significantly improves face verification accuracy, achieving 99.15% on the LFW dataset, a 67% reduction in error rate compared to the best deep learning result on the same dataset. The learned DeepID2 features are generalizable to new identities and tasks, making them highly effective for face recognition. The paper also discusses the importance of balancing the identification and verification signals and the impact of identity information on feature learning. Finally, the authors demonstrate the practicality and efficiency of their method by integrating DeepID2 into a conventional face verification pipeline, achieving state-of-the-art performance.This paper addresses the challenge of developing effective feature representations for face recognition by leveraging deep learning and using both face identification and verification signals as supervision. The authors propose DeepID2, a feature representation learned through deep convolutional networks. The identification task increases inter-personal variations by separating DeepID2 extracted from different identities, while the verification task reduces intra-personal variations by pulling DeepID2 extracted from the same identity together. This approach significantly improves face verification accuracy, achieving 99.15% on the LFW dataset, a 67% reduction in error rate compared to the best deep learning result on the same dataset. The learned DeepID2 features are generalizable to new identities and tasks, making them highly effective for face recognition. The paper also discusses the importance of balancing the identification and verification signals and the impact of identity information on feature learning. Finally, the authors demonstrate the practicality and efficiency of their method by integrating DeepID2 into a conventional face verification pipeline, achieving state-of-the-art performance.