Advances in 3D Gaussian Splatting

The field of 3D Gaussian Splatting (3DGS) is rapidly advancing with a focus on improving rendering quality, efficiency, and scalability. Recent developments have led to the creation of innovative methods for accelerating 3DGS, such as enhancing GPU triangle rasterizers and proposing optimized minimal Gaussian representations. These advancements have resulted in significant reductions in computation time and improvements in rendering quality, making 3DGS a more viable option for real-time applications. Furthermore, researchers have explored the use of 3DGS in novel view synthesis, global illumination, and watermarking, demonstrating its potential in various areas of computer graphics and vision. Notable papers in this area include GauRast, which proposes an acceleration strategy for 3DGS by leveraging the similarities between the 3DGS pipeline and the conventional graphics pipeline in modern GPUs. Optimized Minimal 3D Gaussian Splatting is another noteworthy paper that significantly reduces storage requirements while maintaining high rendering quality. ProtoGS, GaussianFocus, and StableGS also present innovative approaches to improving 3DGS, such as learning Gaussian prototypes, incorporating patch attention algorithms, and eliminating floater artifacts.

Sources

GauRast: Enhancing GPU Triangle Rasterizers to Accelerate 3D Gaussian Splatting

Optimized Minimal 3D Gaussian Splatting

ProtoGS: Efficient and High-Quality Rendering with 3D Gaussian Prototypes

GaussianFocus: Constrained Attention Focus for 3D Gaussian Splatting

Real-time Global Illumination for Dynamic 3D Gaussian Scenes

DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds

StableGS: A Floater-Free Framework for 3D Gaussian Splatting

Hardware-Rasterized Ray-Based Gaussian Splatting

GS-Marker: Generalizable and Robust Watermarking for 3D Gaussian Splatting

TC-GS: Tri-plane based compression for 3D Gaussian Splatting

Built with on top of