Text this: Interactive Co-segmentation of Objects in Image Collections