Yiwu Zhong Qualifying Exam Presentation

Abstract: We proposed a novel image captioning model by exploring the sub-graphs of image scene graph. Our key idea is to identify important sub-graphs, with each sub-graph capturing a semantic component of the input image. By selectively attending to different sub-graphs, our model accounts for diverse, grounded and controllable image captioning at the same time, while maintaining the quality of the captions.

Qualifying exam committee:
Yin Li (advisor)
Jerry Zhu
Yingyu Liang