What is Gaussian splatting?
Gaussian splatting (3DGS) turns a photo into millions of soft 3D points instead of a mesh — the technique that lets you walk through a room in a browser.
3D Gaussian Splatting (3DGS) is a technique for representing a 3D scene as a cloud of millions of small, semi-transparent, colored ellipsoids ("splats") instead of the triangles a traditional 3D mesh uses. Each splat carries a position, size, orientation, color, and opacity; rendering the scene means blending the splats a camera can see, which a GPU can do fast enough for real-time movement.
Why it matters for a photo of your home
Older photo-to-3D techniques either built a rough mesh (blocky, loses detail) or a NeRF (photorealistic, but too slow to render interactively). Gaussian splatting sits in between: it keeps the photorealism of a NeRF while rendering at real-time frame rates in a normal web browser, on a phone included. That combination is what makes free WASD-style movement through a scanned room possible at all, rather than just orbiting a fixed point — the technique produces the data; a renderer like Spark is what turns it into something you can walk through.
How a splat scene gets to your browser
A splat scene is usually distributed as a .spz (compressed) or .ply (uncompressed) file containing the point cloud, then rendered with a WebGL2 library — Spark is the most widely used one for Three.js. Because it's just files and WebGL, no plugin or app install is required.
Related terms
- NeRF (Neural Radiance Field) — an earlier photorealistic scene-reconstruction method; slower to render than Gaussian splatting
- Photogrammetry — building a 3D mesh from many overlapping photos, the technique behind most drone and LiDAR scans
- Point cloud — a set of 3D points in space, the raw data structure Gaussian splatting extends with size, color, and opacity per point