Modeling Dyadic Human Interactions A study of methods for training pose+motion models of fine-grained face-to-face interactions in unsegmented videos