Key points
- 3D Gaussian splatting represents a scene as a set of learnable 3D Gaussians and renders them with a fast differentiable rasteriser, achieving real-time radiance field rendering.
- The method won the SIGGRAPH 2023 Best Paper award and was integrated into compositing and virtual-production pipelines within two years of publication.
- Within two years of publication, 3D Gaussian splatting became an industry-standard technique in virtual production and compositing, adopted in commercial tools and pipeline workflows.
Summary
Kerbl et al. (2023) introduce 3D Gaussian splatting, representing scenes as collections of 3D Gaussians optimised from multi-view photographs and rendered via a differentiable tile-based rasteriser. The approach achieves real-time novel-view synthesis at quality comparable to NeRF-based methods while being orders of magnitude faster to render. SIGGRAPH 2023 Best Paper recognition and rapid adoption in virtual production and compositing pipelines establish this as a shaped-the-field result with direct relevance to animation and VFX education.
Related items
- Instant Neural Graphics Primitives with a Multiresolution Hash Encoding
- Nerfstudio: A Modular Framework for Neural Radiance Field Development
- DifFRelight: Diffusion-Based Facial Performance Relighting
Source
Source: ACM Transactions on Graphics (SIGGRAPH 2023) ↗ (Research)
Cite this item
Kerbl, B., Kopanas, G., Leimkuehler, T. & Drettakis, G. (2023). ‘3D Gaussian Splatting for Real-Time Radiance Field Rendering’, ACM Transactions on Graphics (SIGGRAPH 2023). Available at: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
Your reference manager can also read this page directly: with the Zotero (or Mendeley) browser connector installed, save it straight to your library. Whole-collection exports: RIS, BibTeX, CSL-JSON.