caiyuanhao1998/Open-DiffusionGS

What it solves

Open-DiffusionGS addresses the slow speed and complexity of image-to-3D generation. Traditional methods often rely on multi-stage processes or 2D multi-view diffusion models, which can be computationally expensive and time-consuming. This project provides a single-stage approach for fast 3D object generation and scene reconstruction from a single image.

How it works

The system "bakes" Gaussian Splatting directly into a diffusion denoiser. Unlike other methods, it does not require a 2D multi-view diffusion model or a depth estimator for scene reconstruction. It can generate 3D objects with mesh exportation capabilities and reconstruct scenes in approximately 6 seconds.

Who it’s for

This tool is designed for researchers and developers working in 3D computer vision, generative AI, and digital content creation who need high-speed, single-stage 3D reconstruction and generation.

Highlights

• Single-stage generation: Eliminates the need for multi-view diffusion models.
• High speed: Capable of scene reconstruction in ~6 seconds and object inference in 24 seconds (significantly faster than competitors like Hunyuan-v2.5).
• Versatile output: Supports both single-view 3D object generation and scene-level reconstruction.
• Mesh export: Includes functionality to export generated 3D objects as meshes.