Using Navigational Information to Learn Visual Representations