Learning Shape-Invariant Representation for Generalizable Semantic Segmentation | IEEE Journals & Magazine | IEEE Xplore