How to make Face Recognition Work: The Power of Modeling Context

Face recognition in the wild has been one of the longest standing computer vision challenges. While there has been constant improvement over the years, the variations in appearance, illumination, pose etc. still makes it one the hardest task to do well. In this paper we summarize two techniques that leverage context and show significant improvement over vision only methods. At the heart of the approach is a probabilistic model of context that captures dependencies induced via set of contextual relations. The model allows application of standard variational inference procedures to infer labels that are consistent with contextual constraints.