Click the dots to switch between three vision tasks: say what is in a picture, box where it is, or give every pixel a label.
