Learning Common and Specific Features for RGB-D Semantic Segmentation with Deconvolutional Networks | Read Paper on Bytez