Hirokatsu Kataoka's HP

1. Chief Senior Researcher, AIST/ 2. Academic Visitor, Visual Geometry Group (VGG), University of Oxford / 3. Visiting Associate Professor, Keio University / 4. Adjunct Associate Professor, Tokyo Denki University / 5. Research Advisor, SB Intuitions / 6. Principal Investigator, cvpaper.challenge / 7. Principal Investigator, LIMIT.Lab

As of 2026, he is formulating research themes grounded in the philosophies of “LIMIT” and “VGI.” Within LIMIT, he focuses on building foundation models under limited resources (see also LIMIT.Lab). In parallel, with the advent of the AGI era, he has established the philosophy of Visual General Intelligence (VGI), which aims to advance pure vision research, and is actively pursuing this direction (see also the CVPR 2026 VGI Workshop).

He proposed 3D ResNets as a baseline spatiotemporal model, which became one of the top 0.5% most-cited papers at CVPR over a five-year period. He also introduced Formula-Driven Supervised Learning (FDSL), a synthetic pre-training method without real images and human labor, which earned an ACCV 2020 Best Paper Honorable Mention Award. He received the Fujiwara Prize in 2014 (valedictorian equivalent) from Keio University, participated in the ECCV 2016 Workshop “Brave New Idea,” and won the AIST Best Paper Award in 2019 and 2022, was also a BMVC 2023 Best Industry Paper Finalist. He has served as primary organizer of the LIMIT Workshops at ICCV 2023, CVPR 2024, and ICCV 2025, as Area Chair for CVPR 2024 and 2025, and will serve as an IEEE TPAMI Associate Editor beginning in 2025. His work has been featured in MIT Technology Review.