Many representation and modalities are developed for better scene understanding as images, videos, point clouds, etc. In this thesis, we intentionally characterize scene representation as depth maps in order to leverage ...