云南省人工智能重点实验室

科研成果

当前位置：首页>科研成果>论文列表

论文作者：	Tu Yunbin;Yao Tingting; Li Liang; Lou Jiedong;Gao Shengxiang;Yu Zhengtao; Yan Chenggang
论文名称：	Semantic Relation-aware Difference Representation Learning for Change Captioning
论文发表刊物：	In Proceedings of Findings of the Association for Computational Linguistics(ACL2021)
会议地点：
会议时间：
卷号页码：
论文描述：
收录情况：	EI Indexed
论文摘要：
Change captioning is to describe the differencein a pair of images with a natural languagesentence. In this task,the distractors, suchas the illumination or viewpoint change, bringthe huge challenges about learning the difference representation. In this paper, we proposea semantic relation-aware difference represen-tation learning network to explicitly learn thedifference representation in the existence ofdistractors. Specifcally, we introduce a selfsemantic relation embedding block to explorethe underlying changed objects and design across-semantic relation measuring block to lo-calize the real change and learn the discrimina-tive difference representation. Besides, relyingon the POS of words, we devise an attentionbased visual switch to dynamically use visualinformation for caption generation. Extensiveexperiments show that our method achievesthe state-of-the-art performances on CLEVRChange and Spot-the-Diff datasets.
点击此处下载文章内容