Uncertainty Matters in Dynamic Gaussian Splatting for Monocular 4D Reconstruction

Researchers introduce USplat4D, a new uncertainty-aware dynamic Gaussian Splatting framework that improves 3D scene reconstruction from monocular video by modeling per-Gaussian uncertainty. The approach addresses motion drift and poor synthesis quality by treating well-observed Gaussians as reliable anchors while handling poorly observed ones as less reliable.

• →USplat4D introduces uncertainty modeling to dynamic Gaussian Splatting for better 4D reconstruction from monocular input.
• →The framework estimates time-varying per-Gaussian uncertainty to construct spatio-temporal graphs for optimization.
• →Well-observed Gaussians across views and time serve as reliable anchors to guide motion estimation.
• →The approach reduces motion drift under occlusion and improves synthesis quality at extreme viewpoints.
• →Experiments on real and synthetic datasets show consistent improvements over vanilla dynamic Gaussian Splatting models.