gaussian-splatting: Real-Time 3D Radiance Field Rendering
This repository profile is provided by osrepos.com, an open source repository discovery platform.
Summary
gaussian-splatting is the original reference implementation for real-time radiance field rendering. This repository introduces a novel approach using 3D Gaussians for high-quality, real-time novel-view synthesis at 1080p resolution, offering significant advancements in computer graphics and vision. Developed by GRAPHDECO Inria, it provides a robust framework for 3D scene reconstruction and interactive visualization.
Repository Information
Topics
Click on any tag to explore related repositories
Use at your own risk
OSRepos shares public repositories for knowledge and discovery only. Any installation, execution, configuration, or use of code from these repositories is the user's own responsibility. Always review the repository, source code, dependencies, licenses, and security implications before running or installing anything. OSRepos is not responsible for issues, damages, or losses resulting from third-party repositories.
Introduction
The gaussian-splatting repository presents the original reference implementation of "3D Gaussian Splatting for Real-Time Radiance Field Rendering," a groundbreaking method developed by GRAPHDECO Inria. This innovative approach revolutionizes novel-view synthesis by achieving state-of-the-art visual quality and real-time performance at 1080p resolution. It leverages 3D Gaussians to represent scenes, optimizing anisotropic covariance and employing a fast, visibility-aware rendering algorithm to deliver unparalleled speed and fidelity in 3D reconstruction and rendering.
Installation
To get started with gaussian-splatting, begin by cloning the repository, ensuring to include submodules:
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
The project primarily uses Python with PyTorch and CUDA extensions. A Conda environment is highly recommended for managing dependencies. Hardware requirements include a CUDA-ready GPU with Compute Capability 7.0+ and at least 24 GB VRAM for full training quality, though smaller VRAM configurations are possible with adjustments. Detailed setup instructions for both Windows and Ubuntu Linux, including C++ compiler and CUDA SDK requirements, are provided in the official documentation.
Examples
The gaussian-splatting codebase offers a comprehensive suite of tools for training, rendering, and evaluation:
- Optimizer: Use
python train.py -s <path to dataset>to train 3D Gaussian models from SfM inputs. - Evaluation: Generate renderings and compute error metrics with
python render.pyandpython metrics.pyafter training, optionally using a train/test split with the--evalflag. - Interactive Viewers: Explore your trained models in real-time using the provided network viewer (for live training visualization) and the OpenGL-based real-time viewer. Pre-built Windows binaries are available for convenience.
- Processing Custom Scenes: Convert your own images into optimization-ready SfM datasets using
python convert.py, which integrates with COLMAP and ImageMagick. - New Features: Recent updates include training speed acceleration (using
sparse_adamoptimizer), depth regularization for improved scene reconstruction, exposure compensation for "in the wild" captures, and anti-aliasing with an EWA Filter. The SIBR viewers also feature a "Top View" for visualizing camera placement and an experimental OpenXR branch for VR support.
Why Use
gaussian-splatting stands out for its ability to produce exceptionally high visual quality for novel-view synthesis while maintaining real-time rendering speeds, a significant challenge for previous radiance field methods. Its efficient representation using 3D Gaussians and optimized rendering pipeline make it ideal for applications requiring interactive 3D experiences, such as virtual reality, augmented reality, and advanced computer graphics. The project's active development, robust feature set, and strong academic backing from Inria position it as a leading solution in the field of neural rendering and 3D reconstruction.
Links
- GitHub Repository: https://github.com/graphdeco-inria/gaussian-splatting
- Project Webpage: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
- Full Paper: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/3d_gaussian_splatting_high.pdf
- Video Overview: https://youtu.be/T_kXY43VZnk
- Pre-trained Models: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/datasets/pretrained/models.zip
- Windows Viewers: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/binaries/viewers.zip
- Colab Template: https://github.com/camenduru/gaussian-splatting-colab
- Step-by-step Tutorial: https://www.youtube.com/watch?v=UXtuigy_wYc
Source repository
Open the original repository on GitHub.