Predicting which spatial regions in a scene are occupied by objects or obstacles, providing dense semantic layout.