From Pixels To People: Graph Based Methods For Grouping Problems In Computer Vision