Towards Scene Understanding: Unsupervised Monocular Depth Estimation with Semantic-aware Representation