DisasterM3 Team
The DisasterM3 team members are:
Junjue Wang
Assistant Professor, The University of Tokyo
Junjue is an Assistant Professor at The University of Tokyo. His research focuses on multi-modal remote sensing and vision–language learning for Earth observation. He is passionate about both research and data challenges, and enjoys applying novel algorithms to real-world geoscience problems. He is the lead developer of benchmarks including LoveDA, EarthVQA, and DisasterM3. His team won first place in the 2025 AI4EO “AI for Earthquake Response” challenge and the 2022 Landslide4Sense competition, and second place in the 2019 IEEE GRSS Data Fusion Contest, among other awards.
Weihao Xuan
Ph.D. Student, The University of Tokyo & RIKEN AIP
Weihao is a Ph.D. student at The University of Tokyo and also a Junior Research Associate at RIKEN AIP. His research builds reliable multimodal AI agents for the physical world. He studies how language models, vision-language models, and tool-using agents behave under uncertainty: what they know, when they hallucinate, and how they can verify their own claims. He grounds this work in Earth observation and geospatial intelligence, where models integrate imagery, language, time, location, and domain knowledge to support high-stakes decisions in disaster response, urban understanding, and climate adaptation. More broadly, he extends these principles to scientific domains where reliability and evidence are central to expert decision-making.
Pengyu Dai
Ph.D. Student, The University of Tokyo & RIKEN AIP
Pengyu Dai received the M.Eng. degree from the Institute of Science Tokyo (formerly Tokyo Institute of Technology), Tokyo, Japan. He is currently pursuing the Ph.D. degree with the Department of Information Science and Technology, The University of Tokyo, Tokyo, Japan. His current research interests include remote sensing foundation models, self-supervised learning, and multimodal vision-language models.
Jingjun Yi
Ph.D. Student, University of Amsterdam
Jingjun Yi is a Ph.D. student at the University of Amsterdam. His research focuses on hyperbolic learning in computer vision, large multi-modal models, and remote sensing. He has published papers in leading venues and journals, including NeurIPS, ICCV, ACM MM, AAAI, TPAMI, ISPRS Journal of Photogrammetry and Remote Sensing, and IEEE TGRS. His work explores robust visual representation learning, generalizable dense prediction, image restoration, and multi-modal Earth observation. He received the NeurIPS 2025 Scholar Award and won first place in the 2025 European Space Agency AI for Earthquake Response Challenge.
Naoto Yokoya
Professor, The University of Tokyo & RIKEN AIP
Naoto Yokoya is a Professor at The University of Tokyo and Team Leader at RIKEN AIP. He focuses on developing intelligent systems that understand and reason about the complex physical and semantic structure of the real world from visual and multimodal observations. His work is positioned at the intersection of computer vision, machine learning, and data fusion, aiming to establish fundamental principles of visual and spatial intelligence. Earth observation serves as a particularly challenging and socially important domain where these principles are tested and advanced.
