Jaesik Park

Korea Advanced Institute of Science and Technology, Pohang University of Science and Technology, Seoul National University

Papers

Total Citations

H-Index

About

Jaesik Park is a computer vision and robotics researcher whose work spans 3D scene understanding, sensor calibration, and geometric deep learning. His early contributions addressed foundational challenges in depth sensing, most notably his 2011 work introducing a novel 2.5D calibration pattern for Time-of-Flight and RGB camera fusion systems — a practical solution to extrinsic calibration that has garnered 22 citations and remains relevant as depth cameras proliferate in robotics applications. Complementing this, his 2010 research on depth-cue-guided small object detection demonstrated an early appreciation for leveraging geometric information to overcome the limitations of appearance-based recognition. Park's more recent work reflects a natural evolution toward learned 3D representations. His 2020 research on combinatorial 3D shape generation via sequential primitive assembly tackled the combinatorial complexity inherent in constructing structured geometric blueprints — a problem with direct implications for robotic manipulation and scene reconstruction. His subsequent investigations into point cloud colorization and SO(3)-invariant semantic correspondence push the boundaries of 3D understanding in uncontrolled, real-world environments, addressing critical gaps around rotation robustness and meaningful visualization. Across his career, Park consistently bridges theoretical elegance with practical robotics applicability, making his research particularly valuable for students working at the intersection of 3D vision and intelligent systems.