Title: Enhancing 3D Scene Reconstruction in NeRFs
Date: 2024-03-01 14:20-15:30
Location: CSIE R103
Speaker: Prof. Yu-Lun (Alex) Liu, National Yang Ming Chiao Tung University
Host: Prof. Lung-Pan Cheng
Abstract:
This presentation explores cutting-edge advancements in Neural Radiance Fields (NeRF), tackling key challenges in few-shot learning, dynamic scene reconstruction, and simultaneous camera pose and scene geometry optimization. We introduce a novel approach that accelerates convergence in few-shot scenarios by leveraging multi-scale self-adaptation without relying on pre-trained priors. Additionally, we discuss a method that enhances the robustness of joint camera pose and scene geometry optimization through decomposed low-rank tensorial representations and sophisticated training schedules. The talk also covers a progressive optimization strategy for reconstructing large-scale scenes from casually captured videos, improving robustness and scalability. Finally, we present a solution for dynamic scene modeling that jointly estimates static and dynamic radiance fields along with camera parameters, offering improved performance in dynamic environments. These innovations collectively advance the NeRF field, enabling more efficient, robust, and realistic 3D scene rendering and view synthesis.
Short bio:
Yu-Lun (Alex) Liu is an Assistant Professor in the Department of Computer Science at National Yang Ming Chiao Tung University. His research spans image/video processing, computer vision, and computational photography, focusing on integrating machine learning, geometry, and domain-specific insights. Before his current role, he was a Research Scientist Intern at Meta Reality Labs Research and a Senior Software Engineer at MediaTek Inc. He completed his PhD at NTU, CSIE, where he was a member of CMLab.