Recent state-of-the-art deep learning frameworks require large, fully annotated training datasets that are, depending on the objective, time-consuming to generate. While in most fields, these labelling tasks can be ...