3D Gaussian Splatting for Real-Time Radiance Field Rendering
The foundational paper for 3DGS, breaking the barrier between high-quality view synthesis and real-time rendering.
Authors / Team
Bernhard Kerbl & Georgios Kopanas · Researcher
Location
法国 · 尼斯
Year
2023
Deep Dive
The seminal SIGGRAPH 2023 paper from the Inria team. By explicitly representing the scene with an array of anisotropic Gaussian ellipsoids paired with a specialized differentiable rasterization pipeline, the work achieves unprecedented real-time rendering speeds while matching or surpassing the image quality of state-of-the-art NeRF models. The project page hosts rendering code, training framework, full paper, and benchmarks against classic datasets — the absolute foundation for all modern spatial computing and Gaussian derivative products.
What we learn
- 01
Moving from implicit networks to explicit Gaussian point clouds is the crucial step toward real-time interactive rasterization.
- 02
Anisotropy of Gaussians dramatically enhances the depiction of complex geometry and high-frequency edges.
- 03
Adaptive density control effectively resolves rendering artifacts in both over-fitted and under-fitted regions.
Verbatim quote
"Radiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or videos."— source ↗
Tags
Links
Sources