3DWM @ ECCV 2026
M.S. Student, KAIST CVLAB
Seonghu Jeon.
I build geometry-aware generative models for 3D/4D vision and robot learning. My work connects multi-view generation, geometric foundation models, and visuomotor policy learning.
Featured research
News & notes
2026.08
GAM was accepted to 3DWM @ ECCV 2026.
2026.07
GLD was selected for a Long Oral Presentation at ECCV 2026.
2026.07
Joined MAUM AI's WoRV Division, Future Intelligence Lab as a research intern for Robotics Foundation Model development.
2026.03
Joined KAIST CVLAB as an M.S. student under Prof. Seungryong Kim.
2026.02
CAMEO accepted to CVPR 2026 for correspondence-attention alignment in multi-view diffusion.
2026.02
Graduated from Korea University, B.S. in Computer Science & Engineering, with Great Honors.
Focus
01
3D / 4D vision
Reconstructing static and dynamic scenes from sparse views, and studying how geometry-trained foundation models transfer to view synthesis, including GLD, an ECCV 2026 Long Oral Presentation.
02
Generative models
Diffusion and flow matching with structured conditioning across correspondence, geometry, and motion. Architecture work and better source distributions.
03
Robotics
The destination. If a robot can imagine a scene's geometry forward in time, it can plan in it, and that loop closes through generative models.
Experience
01
Research Intern
Participating in Robotics Foundation Model development.
02
M.S. Student
Advised by Prof. Seungryong Kim. Multi-view diffusion, geometric foundation models, 4D scene generation.
03
Research Intern
Pre-graduate research on motion transfer and correspondence-attention. Multiple co-authored papers from this period.
04
B.S., Computer Science & Engineering
Graduated with Great Honors (4.46 / 4.50). Coursework in vision, graphics, and deep learning.