Bolt3D enables rapid generation of 3D scenes from input images using a multi-view diffusion model and Geometry VAE. It can handle varying numbers of input views and generates high-quality 3D reconstructions without additional reprojection or inpainting mechanisms. This method outperforms other techniques by providing generative capabilities with reduced inference costs.

2m read timeFrom szymanowiczs.github.io
Post cover image
Table of contents
Geometry VAE

Sort: