About graphdeco-inria/gaussian-splatting

The official implementation of 3D Gaussian Splatting for Real-Time Radiance Field Rendering (Kerbl et al., 2023). See the paper page for context.

This repository introduced the reference CUDA kernels for differentiable Gaussian rasterization, adaptive density control, and the spherical harmonics view-dependent color model that essentially every downstream 3DGS method builds on. It also ships the training pipeline that takes a sparse COLMAP SfM point cloud and densifies it into millions of anisotropic 3D Gaussians. The Tesserae demo corpus mirrors only the README to demonstrate cross-graph linking between papers and their canonical implementations; the upstream license is non-commercial research only, so the code itself is not redistributed here.