Google presents CAT4D Create Anything in 4D with Multi-View Video Diffusion Models

discuss:

Thanks for sharing our work! Project page: arXiv:

The paper presents a method called CAT4D (Create Anything in 4D) that can generate high-quality dynamic 3D scenes from a single input monocular video. The key idea is to leverage a multi-view video diffusion model trained on a diverse combination of datasets to enable novel view synthesis at any specified camera poses and timestamps. The authors evaluate their method on various tasks, including novel view synthesis, sparse-view static 3D reconstruction in the presence of scene motion, and 4D reconstruction from monocular videos. They show that their method can generate high-quality dynamic 3D scenes and outperforms existing state-of-the-art models that depend on multiple priors and external sources of information. full paper:

CAT4D? More like create anything in 4D and amaze me!

4D? Like 4 dimensions? If so - this is not it, this is 3D.

Any chance of a code release?

I wish my cat could bake like that. jk I don't have a cat

oooo

👀