Predicting which 3D spatial locations are occupied and what semantic class (car, pedestrian, etc.) occupies them.