gaussian-splatting

Why we included this project

3D Gaussian Splatting is the method that made high-quality novel-view synthesis from ordinary photos practical at real-time frame rates, and this repo is the code its authors released alongside the paper. For anyone doing 3D reconstruction or scene capture, it is the reference point most radiance-field tools are compared against, and the usual starting place before moving to faster or more specialized forks. The repo bundles a PyTorch optimizer that turns structure-from-motion point clouds into a 3D Gaussian scene model, an OpenGL viewer for rendering trained models in real time, and a script that prepares your own image sets for training. It is research code with real hardware demands: a CUDA GPU with 24 GB VRAM for full-quality training. That makes it a fit for developers and researchers who want the algorithm's actual behavior rather than a turnkey product. Because this is the release from the paper's own authors, it is also the most dependable reference for understanding how the technique works and for validating other implementations.