Deformable Gaussian Occupancy: Decoupling Rigid and Nonrigid Motion with Factorized Distillation

Researchers introduced Deformable Gaussian Occupancy (DeGO), a weakly supervised 3D occupancy framework for autonomous driving.

DeGO combines decoupled Gaussian deformation with factorized 4D distillation from the VGGT foundation model to capture both rigid and deformable motion.

On Occ3D-NuScenes, the method reports state-of-the-art weakly supervised performance, with strong gains in human-centric and overall accuracy.

The work could improve dynamic scene understanding and safer tracking of people and other deformable agents in driving environments.