Deep Face Recognition: A Survey Deep learning has significantly improved face recognition (FR) performance since its introduction in 2014 with the DeepFace and DeepID breakthroughs. This survey provides a comprehensive review of recent developments in deep FR, covering algorithm designs, databases, protocols, and application scenarios. It summarizes different network architectures and loss functions, categorizes face processing methods into "one-to-many augmentation" and "many-to-one normalization," and compares commonly used databases for training and evaluation. It also reviews various FR scenarios, such as cross-factor, heterogeneous, multiple-media, and industrial applications. The survey highlights technical challenges and promising directions in deep FR. Face recognition has been a long-standing research topic in the CVPR community. Early methods, such as Eigenface, used holistic approaches to derive low-dimensional representations. However, these methods struggled with uncontrolled facial changes. Local-feature-based methods, such as Gabor and LBP, were introduced in the early 2000s, but they lacked distinctiveness and compactness. Learning-based local descriptors were introduced in the early 2010s, but shallow representations still had limitations in robustness against complex facial variations. Deep learning methods, such as convolutional neural networks (CNNs), use multiple layers of processing units for feature extraction and transformation. They learn multiple levels of representations that correspond to different levels of abstraction, showing strong invariance to face pose, lighting, and expression changes. Deep learning has reshaped the research landscape of FR in almost all aspects, including algorithm designs, training/test datasets, application scenarios, and evaluation protocols. The survey provides a systematic review of the evolution of network architectures and loss functions for deep FR. Various loss functions are categorized into Euclidean-distance-based loss, angular/cosine-margin-based loss, and softmax loss and its variations. Both mainstream network architectures, such as DeepFace, DeepID series, VGGFace, FaceNet, and VGGFace2, and other architectures designed for FR are covered. The survey categorizes new face processing methods based on deep learning, such as those used to handle recognition difficulty on pose changes, into two classes: "one-to-many augmentation" and "many-to-one normalization," and discusses how emerging generative adversarial network (GAN) facilitates deep FR. It presents a comparison and analysis on public available databases that are of vital importance for both model training and testing. Major FR benchmarks, such as LFW, IJB-A/B/C, Megaface, and MS-Celeb-1M, are reviewed and compared, in terms of the four aspects: training methodology, evaluation tasks and metrics, and recognition scenes, which provides an useful reference for training and testing deep FR. Besides the general purpose tasks defined by the major databases, the survey summarizes a dozen scenario-specific databases and solutions that are still challenging for deep learning, such as anti-attack, cross-pose FR, and cross-age FR. ByDeep Face Recognition: A Survey Deep learning has significantly improved face recognition (FR) performance since its introduction in 2014 with the DeepFace and DeepID breakthroughs. This survey provides a comprehensive review of recent developments in deep FR, covering algorithm designs, databases, protocols, and application scenarios. It summarizes different network architectures and loss functions, categorizes face processing methods into "one-to-many augmentation" and "many-to-one normalization," and compares commonly used databases for training and evaluation. It also reviews various FR scenarios, such as cross-factor, heterogeneous, multiple-media, and industrial applications. The survey highlights technical challenges and promising directions in deep FR. Face recognition has been a long-standing research topic in the CVPR community. Early methods, such as Eigenface, used holistic approaches to derive low-dimensional representations. However, these methods struggled with uncontrolled facial changes. Local-feature-based methods, such as Gabor and LBP, were introduced in the early 2000s, but they lacked distinctiveness and compactness. Learning-based local descriptors were introduced in the early 2010s, but shallow representations still had limitations in robustness against complex facial variations. Deep learning methods, such as convolutional neural networks (CNNs), use multiple layers of processing units for feature extraction and transformation. They learn multiple levels of representations that correspond to different levels of abstraction, showing strong invariance to face pose, lighting, and expression changes. Deep learning has reshaped the research landscape of FR in almost all aspects, including algorithm designs, training/test datasets, application scenarios, and evaluation protocols. The survey provides a systematic review of the evolution of network architectures and loss functions for deep FR. Various loss functions are categorized into Euclidean-distance-based loss, angular/cosine-margin-based loss, and softmax loss and its variations. Both mainstream network architectures, such as DeepFace, DeepID series, VGGFace, FaceNet, and VGGFace2, and other architectures designed for FR are covered. The survey categorizes new face processing methods based on deep learning, such as those used to handle recognition difficulty on pose changes, into two classes: "one-to-many augmentation" and "many-to-one normalization," and discusses how emerging generative adversarial network (GAN) facilitates deep FR. It presents a comparison and analysis on public available databases that are of vital importance for both model training and testing. Major FR benchmarks, such as LFW, IJB-A/B/C, Megaface, and MS-Celeb-1M, are reviewed and compared, in terms of the four aspects: training methodology, evaluation tasks and metrics, and recognition scenes, which provides an useful reference for training and testing deep FR. Besides the general purpose tasks defined by the major databases, the survey summarizes a dozen scenario-specific databases and solutions that are still challenging for deep learning, such as anti-attack, cross-pose FR, and cross-age FR. By