Title：Recent Results on 3D Vision in Dynamic Scenes
Date：2025/03/28 15:30-17:00
Location: R103, CSIE
Speakers: Prof. Ming-Hsuan Yang
Host: 莊永裕教授
Abstract:
Abstract: Estimating geometry from dynamic scenes, where objects move and deform over time, remains a fundamental challenge in computer vision. Existing methods often depend on multi-stage pipelines or global optimization techniques that break down the problem into subtasks, such as depth estimation and optical flow, leading to complex systems susceptible to error propagation. In this talk, I will first introduce MonST3R, a novel geometry-first approach that directly estimates per-timestep geometry from dynamic scenes. Next, I will present NoPoSplat, a feed-forward model that reconstructs 3D scenes parameterized by 3D Gaussians from sparse, unposed multi-view images.
Additionally, I will discuss FaceLift, an innovative feed-forward approach for rapid, high-quality 360-degree head reconstruction from single images. If time permits, I will highlight our Gaga framework, which reconstructs and segments open-world 3D scenes using inconsistent 2D masks predicted by zero-shot, class-agnostic segmentation models.
Biography:
Ming-Hsuan Yang is a Professor at the University of California, Merced, and a Research Scientist at Google DeepMind. He has received numerous awards, including the Google Faculty Award (2009), the NSF CAREER Award (2012), and the Nvidia Pioneer Research Award (2017, 2018). He earned the Best Paper Honorable Mention at UIST 2017, CVPR 2018, and ACCV 2018, as well as the Longuet-Higgins Prize for Test-of-Time at CVPR 2023, Best Paper at ICML 2024, and the Test-of-Time Award at WACV 2025. Yang is an Associate Editor-in-Chief of IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), an Associate Editor for the International Journal of Computer Vision (IJCV) and Transactions on Machine Learning Research (TMLR). Previously, he was the Editor-in-Chief of Computer Vision and Image Understanding (CVIU) and Program Co-Chair for ICCV 2019. He is a Fellow of IEEE, ACM, and AAAI.