Task Specific Networks for Identity and Face Variation

Pose and illumination variations are considered as two main challenges that face recognition system encounters. Most existing methods perform face normalization, aiming at untangling identity representation from these variations to improve recognition accuracy. Taking into account face variation representations, this paper proposes Task Specific Networks for the two representations with two novelties. First, we rotate and normalize face image to multi-pose view for one subtask, and learn face variation representations for another. Second, we learn face variation representations in an unsupervised way, which is more robust and more universal. We couple these two representations in the part of reconstructing the original face, where the two representations effect and restrict each other. Extensive experiments demonstrate the superiority of our method in both learning representations and rotating non frontal face image.