Due to computational resource limitations, we were unable to process and train our model on the full Objaverse dataset, only using 5% of it.
A significant limitation of our work is the lack of consideration given to object texture modeling, which could enhance the generation process.
The application of strong 2D priors from recent diffusion models could improve the efficiency and quality of 3D generation methods, marking a promising direction.
Exploring the scaling properties of our method in relation to recent foundation models may unlock new potentials in performance and effectiveness.
Collection
[
|
...
]