Referring Expression Object Segmentation with Caption-Aware Consistency
2019·Arxiv