Description
Image-to-3D usually generates multi-view images first and then reconstructs, a slow, inconsistent two-stage process. DiffusionGS bakes Gaussian splatting into the diffusion denoiser for single-stage image-to-3D generation and reconstruction.
It's fast, scalable and works for objects and scenes; published at ICCV 2025.
Single stage:One step.
Gaussians:Direct 3D.
Objects and scenes:General.
Fast:Scalable.
It's fast, scalable and works for objects and scenes; published at ICCV 2025.
Features
Single stage:One step.
Gaussians:Direct 3D.
Objects and scenes:General.
Fast:Scalable.
