[Abstract]

Recent advances in 3D vision and graphics have enabled remarkable progress in reconstructing and generating high-fidelity 3D scenes. However, existing methods often struggle to balance speed, efficiency, and robustness, which are key factors for real-world deployment. In this talk, I will discuss recent developments that address these challenges, with a focus on 3D Gaussian Splatting as a powerful representation for neural rendering. I will introduce a compact 3D Gaussian representation that enhances efficiency while preserving visual fidelity, followed by techniques for robust 3D modeling that improve reconstruction quality under real-world challenging conditions, such as camera blur. Finally, I will present feed-forward approaches to novel view synthesis that significantly accelerate 3D reconstruction.

[Biography]

Eunbyung Park is an assistant professor in the Department of Artificial Intelligence at Yonsei University, South Korea. Eunbyung Park obtained his B.S. degree in computer science from Kyung Hee University in 2009, his M.S. degree in computer science from Seoul National University in 2011, and his Ph.D. degree in computer science from the University of North Carolina at Chapel Hill in 2019. Before joining Yonsei University, he was an assistant professor at SKKU, a research scientist at Nuro and an applied scientist at Microsoft. His current research interests include 3D vision and generative modeling.