Points say “where.”
Each sample stores a location in 3D. A point cloud has no inherent edges or faces. More samples describe finer detail, but a surface must still be inferred.
Spatial intelligence / Geometry lab
Matter, sampled.
One surface. Three ways to describe it. Explore how a cloud of coordinates becomes connected geometry—or a field of soft, overlapping light.
01 / POINT CLOUD
DRAG TO ORBIT · ARROW KEYS TO ROTATE · SPACE TO PAUSE
Meshes say “connected.”
Triangles connect neighboring vertices into an explicit surface. This view exposes that connectivity as a wireframe: each four-corner patch becomes two triangles.
Splats say “contribute.”
Soft elliptical footprints overlap and blend. Real 3D Gaussian splatting learns positions, covariances, opacity and view-dependent color from photographs.
What this experiment does—and does not—show.
All three views use the same procedurally sampled torus. The splats here are illustrative, screen-space Gaussian ellipses composited back to front. They are not trained 3D Gaussians: this demo does not project learned 3D covariance, optimize against photographs, or reproduce a production 3DGS renderer. The wireframe reveals topology rather than rendering an opaque surface.