Timezone: »

 
Open Visual Knowledge Extraction via Relation-Oriented Multimodality Model Prompting
Hejie Cui · Xinyu Fang · Zihan Zhang · Ran Xu · Xuan Kan · Xin Liu · Manling Li · Yangqiu Song · Carl Yang

Images contain rich relational knowledge that can help machines understand the world. Existing methods on visual knowledge extraction often rely on the pre-defined format (e.g., sub-verb-obj tuples) or vocabulary (e.g., relation types), restricting the expressiveness of the extracted knowledge. In this work, we take a first exploration to a new paradigm of open visual knowledge extraction. To achieve this, we present OpenVik which consists of an open relational region detector to detect regions potentially containing relational knowledge and a visual knowledge generator that generates format-free knowledge by prompting the large multimodality model with the detected region of interest. We also explore two data enhancement techniques for diversifying the generated format-free visual knowledge. Extensive knowledge quality evaluations highlight the correctness and uniqueness of the extracted open visual knowledge by OpenVik. Moreover, integrating our extracted knowledge across various visual reasoning applications shows consistent improvements, indicating the real-world applicability of OpenVik.

Author Information

Hejie Cui (Emory University)

I am a second-year Ph.D. student in Computer Science at Emory University, currently working with Dr. Carl Yang in Emory Graph Mining Lab. I have also been working closely with Dr. Eugene Agichtein in Emory Intelligent Information Access Lab (IR Lab). Before joining Emory, I got my bachelor’s degree in Software Engineering from Tongji University, where I was working with Dr. Lin Zhang. My current research interests lie in graph data mining and structured information systems.

Xinyu Fang (Tongji University)
Zihan Zhang (Tongji University)
Ran Xu (Emory University)
Xuan Kan (Emory University)
Xin Liu (Hong Kong University of Science and Technology)
Xin Liu

I am a fourth-year Ph.D. candidate at CSE Department, the Hong Kong University of Science and Technology, supervised by Prof. Yangqiu Song. My research focuses on pattern mining and learning in graphs and text, and I am interested in graph matching & reasoning, and natural language understanding. I am on the job market for postdoc and industry research positions starting Winter 2022.

Manling Li (University of Illinois at Urbana-Champaign)
Yangqiu Song (Hong Kong University of Science and Technology)
Carl Yang (Emory University)

More from the Same Authors