Fig. 2
From: Generation driven understanding of localized 3D scenes with 3D diffusion model
Local cube. As illustrated in the accompanying figure, a number of local cubes with a volume size of 5%-25% of the original object boundary volume are randomly selected on the 3D object.
