RelationNet: Learning Deep-Aligned Representation for Semantic Image Segmentation | IEEE Conference Publication | IEEE Xplore