Overall framework
Abstract
BulletTime introduces a 4D-controllable video diffusion framework that decouples scene dynamics from camera motion. It conditions generation on continuous world-time sequences and camera trajectories, enabling fine-grained control over both timing and viewpoint.
Type
Publication
In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Tong WU 吴桐
Assistant Professor @ Fudan
My research interests include 3d vision, long-tailed recognition, and robustness.