Compositional models and Structured learning for visual recognition