3D Gaussian Splatting

Conventional documentation often relies on photographs, videos, or static drawings that provide limited spatial context. It can be difficult to communicate the actual appearance, spatial relationships, and visual details of an existing building or site using 2D documentation alone. This can make site review, visual inspection, presentation, and documentation less intuitive. 3D Gaussian Splatting addresses this by transforming captured visual data into an interactive, photorealistic 3D scene that represents the existing environment from multiple viewpoints.

3D Gaussian Splatting is created by capturing the existing environment from multiple viewpoints. The captured images are then processed to estimate camera positions and generate an initial sparse 3D representation. This information is transformed into 3D Gaussian primitives, each defined by its position, scale, orientation, color, and opacity. The Gaussian primitives are then optimized to accurately represent the captured scene, producing a photorealistic 3D visualization that can be viewed from different perspectives.