Support · Datasets
Sample datasets
Every process in Neurones 3D can be tried without data of your own. The demo scenes are generated inside the app on the spot, so there is nothing to download.

Built into the app
Open . Each scene is generated when you pick it.
| Scene | Good for trying |
|---|---|
| Indoor room (point cloud) | Indoor classification, decomposition, meshing, measurement |
| Terrain city (point cloud) | Ground filter, terrain products, the Outdoor model |
| Powerline corridor (point cloud) | The Powerlines model and instances, height above ground |
| Streetscape (point cloud) | The Streetscape model |
| Factory pipework (point cloud) | Cylinders in the decomposition |
| Sculpture (mesh) | Mesh display, conversion to a point cloud |
Under Two clouds at once:
- Two surveys of one site (compare & detect change): a 2024 and a 2026 survey with a building demolished and one built, a 120 mm and a 10 mm settlement patch, and a stockpile grown by 1.5 m, with 8 mm of noise, so the level of detection is about 15 mm. Try M3C2 on it, and a volume on the stockpile.
- Full scan + labelled subsample (transfer a column): a one-in-twenty subsample carrying a classification, to carry across to the full scan.
Public data the numbers were measured on, and data credits
Most figures on this site were measured on public data anyone can obtain, so they can be checked.
- S3DIS, Area 4: an indoor laser scan of about 43.5 million points from the Stanford 3D Indoor Scene Dataset. The indoor classifier, the decomposition, the scene graph and the floor plan figures use it, and so do many screenshots on this site. Source: I. Armeni, S. Sax, A. R. Zamir and S. Savarese, Joint 2D-3D-Semantic Data for Indoor Scene Understanding, arXiv:1702.01105 (2017), Stanford University. Available from its publishers under their own terms; Neurones 3D is not affiliated with or endorsed by them.
- IGN LiDAR HD: France's national aerial LiDAR programme, published openly by the Institut national de l'information géographique et forestière. The ground filter's F1 of 90.3 % was measured against a tile's own production ground class, and the Outdoor model's 86.2 % against held-out tiles. Source: IGN, LiDAR HD, published under the Licence Ouverte / Open Licence 2.0 (Etalab); the renders on this site are derived from it.
- AHN4: the Netherlands' national aerial LiDAR survey (Actueel Hoogtebestand Nederland), published as open data (CC0) via PDOK; used for the Dutch city renders on the features page.
Your own data
Anything in the format matrix opens by drag and drop. Nothing is uploaded: files are read, processed and written on your machine.