Abstract: Multi-class semantic segmentation remains a corner-stone challenge in computer vision. Yet, dataset creation remains excessively demanding in time and effort, especially for specialized ...