[Paper] [Project] [4D Dataset] [BibTeX]
vividzoo-video.mp4
While diffusion models have shown impressive performance in 2D image/video generation, diffusion-based Text-to-Multi-view-Video (T2MVid) generation remains underexplored. The new challenges posed by T2MVid generation lie in the lack of massive captioned multi-view videos and the complexity of modeling such multi-dimensional distribution. To this end, we propose a novel diffusion-based pipeline that generates high-quality multi-view videos centered around a dynamic 3D object from text. Specifically, we factor the T2MVid problem into viewpointspace and time components. Such factorization allows us to combine and reuse layers of advanced pre-trained multi-view image and 2D video diffusion models to ensure multi-view consistency as well as temporal coherence for the generated multi-view videos, largely reducing the training cost. We further introduce alignment modules to align the latent spaces of layers from the pre-trained multi-view and the 2D video diffusion models, addressing the reused layers’ incompatibility that arises from the domain gap between 2D and multi-view data. To facilitate this research line, we further contribute a captioned multi-view video dataset. Experimental results demonstrate that our method generates high-quality multi-view videos, exhibiting vivid motions, temporal coherence, and multi-view consistency, given a variety of text prompts.
[09/27/2024] Accepted to NeurIPS 2024!
[06/14/2024] We have released the paper!
[06/17/2024] We have released our 4D Dataset!
If you would like to download our 4D Dataset data, please fill out this google form, once accepted, we will send you the link to download the data. You should receive the link within one or two days.
@misc{li2024vividzoo,
title={Vivid-ZOO: Multi-View Video Generation with Diffusion Model},
author={Bing Li and Cheng Zheng and Wenxuan Zhu and Jinjie Mai and Biao Zhang and Peter Wonka and Bernard Ghanem},
year={2024},
eprint={2406.08659},
archivePrefix={arXiv},
}