Given sparse unposed views, we leverage rich priors embedded in multiview diffusion models to predict their poses and reconstruct the 3D shape.
Oct 1, 2024
