Spatial AI for the Physical World
3D/4D World Modeling · Efficient Vision-Language Models · Embodied Intelligence
CVSP Lab at Pusan National University develops spatial multimodal AI for understanding, generating, and interacting with real-world environments.
Our research focuses on:
•
3D/4D World Modeling — Gaussian splatting, dynamic scene reconstruction, and immersive media
•
Efficient Vision-Language Models — token pruning, model compression, and reliable multimodal reasoning
•
Embodied Spatial Intelligence — real-world perception and decision-making for robots and media
