Autodistill is an ecosystem for using big, slower foundation models to train small, faster supervised models. Using autodistill and its associated packages, you can go from unlabeled images to inference on a custom model running at the edge with no human intervention in between.