About dreamgaussian/dreamgaussian

The official implementation of DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation (Tang et al., 2023). See the paper page for context.

DreamGaussian replaces the NeRF backbone used by DreamFusion with a 3D Gaussian Splatting representation, then distills a Zero-1-to-3 image-conditioned diffusion prior into it via SDS before extracting a textured mesh. The result is a ~2-minute image-to-3D pipeline that opened the Diffusion-based 3D Generation section of the corpus to feed-forward Gaussian generators like LGM and AGG. Mirrored under MIT — README only.