Recognition of facial attributes using multi-task learning of deep networks

Face recognition is one of important topics in pattern recognition field. Besides recognizing personal identity, there have been numerous studies on recognizing various facial attributes such as gender, age, race, and expression. Recently, rapid growth of deep learning techniques is leading to remarkable improvement of face recognition performances. However, facial attribute recognition is still challenging due to variety of the attributes that can be defined for human faces. As a preliminary work for efficient recognition of various facial attributes, we investigate the effect of multi-task learning of deep neural networks according to diverse combination of different attributes. Through computational experiments on recognizing six attributes by multi-task learning of convolutional neural networks, we show that the effectiveness of multi-task learning is related to the conceptual relationship among attributes, and propose a proper combination of attributes for multi-task learning of facial attribute recognition.