Autodistill

Use big, slow foundation vision models to train smaller, faster models.