Glossary · Processes
Gaussian splatting
A way of representing 3D scans as many small, softly fading blobs of colour. It looks very realistic on screen but is not a model that can be printed directly.
Explained in more detail
Gaussian splatting calculates a scene from many photos or a video, made up of a very large number of small, translucent ellipsoids, known as 3D Gaussians. Each has a position, shape, orientation, colour and opacity. As in photogrammetry, the starting point is the camera positions calculated from the images and a sparse point cloud. The method was presented in 2023 at SIGGRAPH by Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler and George Drettakis of Inria and the MPI Informatik. The result looks highly photorealistic and, according to the authors, runs in real time. It does not produce a closed surface of triangles, however, so you need a mesh for printing.
Where does the term come from?
Named after the Gaussian bell curve, which describes how each blob fades out softly towards its edge, and the English splat. In computer graphics, splatting is a technique in which points are thrown onto the image as small blobs.
In practice
- According to the provider, Scaniverse can save Gaussian splats as well as meshes. For the printer, choose the mesh export.
- The slicer needs a closed triangle mesh as STL, OBJ or 3MF. It cannot open splats.
- Splats are suitable for showing and sharing objects or rooms on screen, not for measuring or designing.
Not to be confused
Like photogrammetry, Gaussian splatting uses photos and calculated camera positions, but the result is not a triangle mesh but a collection of soft blobs. A point cloud is also something different: its points have no extent and no opacity.
Frequently asked questions
Can I 3D print a Gaussian splat?
Not directly. The slicer needs a closed triangle mesh. Some apps additionally calculate a mesh from the same captures, which you export as STL or OBJ.
Why does a splat look more real than a mesh?
Each splat carries its own colour and opacity, so transitions, fine structures and lighting mood look very natural on screen. It says little about the shape of the object, though.