“…Deep learning-based approaches proved to be able to become generalist and robust [27,17,20], but training those approaches is still hampered by the scarcity of publicly available annotated image data sets. In order to overcome this issue, either strong augmentations can be introduced to the image data at arbitrary points within the processing pipeline [36,26,12,35] or synthetic image data can be generated to enrich existing data sets [11,2,4,31,28,30].…”