The unification of semantic and instance segmentation. It assigns a unique label to every pixel, covering both countable 'things' (cars) and amorphous 'stuff' (sky).
The 'holy grail' of scene understanding.
Provides the most complete scene description.