PhD student Wei Cao and Assistant Professor Yaoyao Liu received a Best Paper Award at the 4th Workshop on Generative Models for Computer Vision, which was held during the 2026 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). The paper, "FreeOrbit4D: Training-Free Arbitrary Camera Redirection for Monocular Videos via Foreground-Complete 4D Reconstruction," was co-authored with collaborators from the University of Illinois Urbana-Champaign, the University of Pennsylvania, and Eyeline Labs.
According to the researchers, FreeOrbit4D addresses a challenging problem in visual computing: how to transform an ordinary single-camera video into a new video of the same dynamic scene viewed from a user-specified camera path. This capability could make free-viewpoint replay, cinematic camera movement, and "bullet-time" effects possible without the expensive multi-camera studio setups typically required for such results. The team's method builds a 4D representation of a scene, capturing both its 3D structure and how it changes over time.
"By separating the moving foreground from the background and reconstructing portions of the foreground object that were not visible in the original video, FreeOrbit4D provides stronger geometric guidance for video generation," said Cao. "This allows the system to produce videos that remain visually faithful and temporally consistent, even when the virtual camera moves far away from the original viewpoint."
The paper has been accepted to ACM SIGGRAPH 2026, one of the premier international conferences in computer graphics and interactive techniques.
Led by Liu, the Computer Vision and Machine Learning Group studies a wide range of problems in visual understanding, data-efficient learning, and intelligent perception. The group's research lies at the intersection of computer vision and machine learning, with a special focus on building intelligent visual systems that are continual and data efficient.