Segment images using texts, points, or everything mode