Existing learning-based solutions to medical image segmentation have two important
shortcomings. First for most new segmentation tasks a new model has to be trained or fine …