Gaussian Splatting (3DGS): 3D generator — what it does and what you need to run it

3D Gaussian Splatting (3DGS) is a 3D generator that reconstructs a scene from a set of photos and lets you view it in real time. The tool was created by research groups at Inria and MPII and is distributed as open source.

What it does

As described by the author, the project is a reference implementation of the method "3D Gaussian Splatting for Real-Time Radiance Field Rendering." The tool trains on input images and builds a scene representation that can then be rendered at high speed. The repository lists the following tags: computer-graphics, computer-vision, and radiance-field.

What you need to run it

• Platform: as described by the author, CUDA is required. The optimizer uses PyTorch and CUDA extensions in a Python environment.
• VRAM: as described by the author, training the model to the quality reported in the paper requires 24 GB of VRAM. The authors do not specify memory requirements for inference.
• Code language: Python.
• Code license: non-standard. The authors recommend checking the terms manually.
• Repository: github.com/graphdeco-inria/gaussian-splatting
• Project website: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

Who it's for

The tool is aimed at researchers and developers in computer graphics and computer vision who are willing to work with Python code and have access to a GPU with CUDA support and a large amount of VRAM. Training to the quality described in the original paper requires 24 GB of VRAM, which limits the range of available consumer graphics cards. Those looking for a ready-made application with a graphical interface or lacking suitable hardware will not find this tool suitable.

The repository code was last updated on October 17, 2025, and the repository itself was created in July 2023. Before using it, you should carefully review the non-standard license to understand whether it is acceptable for your project.