Deep Facial Attribute Detection in the Wild: From General to Specific

Accurate facial attribute interpretation is a challenging task in real life due to large head poses, occlusion and illumination variations. This work proposes a general-tospecific deep convolutional network architecture for predicting multiple attributes from a single image in the wild. First, we model the interdependencies of local facial regions by joint learning of all the attributes. Second, task-aware learning is established to explore the disparity regarding each attribute. Finally, an attribute-aware face cropping scheme is proposed to extract more discriminative features from where a certain attribute naturally shows up. The proposed general-to-specific learning strategy ensures both robustness and performance of our model. Extensive experiments on the CelebA and LFWA datasets demonstrate the effectiveness of our architecture and the superiority to state-of-the-art alternatives.