Closing the gap between unconditional and conditional image generation through self-supervised representations.
