Method and System for Semantically Segmenting Scenes of a Video Sequence