Gaussian Surfels

Gaussian surfels, essentially flattened 3D Gaussian points aligned with surface normals, are emerging as a powerful representation for surface reconstruction in various applications. Current research focuses on leveraging Gaussian surfels within models for dynamic scene reconstruction from video, including methods that handle non-rigid motion and complex lighting conditions, and for large-scale tasks like road surface mapping. This approach offers advantages in speed and rendering quality compared to traditional mesh-based methods, impacting fields such as computer graphics, autonomous driving, and virtual avatar creation. The ability to efficiently represent and manipulate surfaces with Gaussian surfels is driving advancements in high-fidelity 3D and 4D modeling.