UniScene3D learns transferable 3D scene representations from multi-view colored pointmaps, unifying RGB appearance and world-aligned geometry within a single ViT encoder. We evaluate its effectiveness ...
3D-Fixer proposes a novel In-Place Completion paradigm to create high-fidelity 3D scene from a single image. Specifically, 3D-Fixer extends 3D object generative priors to generate complete 3D assets ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results