Understand the scene
Unsupervised segmentation
Break the scene into coherent surface patches (walls, floors, table tops, pipe runs) with no training data and no parameters to tune, and get a segment column for every point.
Segmenting is free. Exporting the segment column is Pro.
Part of Point cloud classification, Free point cloud software, Point cloud segmentation.
When to use it
As the first step of understanding a scene, and before classification, which labels segments rather than loose points. Use the Fine preset when borders matter, for example between a wall and a door frame. Segments also feed Mesh Segments (one flat polygon per planar segment) and the labelling grid.
Step by step
- Open Process › Unsupervised Segmentation, or open Properties › AI Analysis and press Segment scene under Unsupervised segmentation.
- Tick Fine (sharp borders) first if you need tight borders between adjacent surfaces; leave it off for the default.
- On a very wide cloud, keep Full detail (tiled) on so every point is segmented rather than a decimated subset.
- Wait for the status bar:
Segmenting… x%, thenSegmented into n segments.The Segments point view comes on, one colour per segment. - Open the segment size histogram under Knowledge to see the distribution; click it for the sortable segment table.
- If the status warns that the segmentation is degenerate (one segment holds most of the cloud), try Fine, or crop to a smaller extent, and Re-run.
What it was measured at
Region growing alone breaks a 43.5 million point building into 9,626 regions; the primitive decomposition built on it reduces the same building to 1,501 reviewable regions in 5.8 s (see decomposition).
From the command line
n3d-render-cli seg-stats --input scan.las # segment count, sizes, degeneracy
n3d-render-cli seg-stats --input scan.las --fine
n3d-render-cli render --input scan.las --segment --out segments.png
n3d-render-cli seg-mesh --input scan.las --out seg_mesh.glb # segments to a mesh (Pro)Limits, stated
- Segments are geometric, not semantic: a segment says this is one surface, not this is a wall. Classification says what it is.
- When a cloud is too large to segment in memory, N3D offers to analyse it at reduced density and says so in a persistent note; labels made that way cannot be re-bound to the full file in a project.
Related questions
Which classifiers ship, and how accurate are they?
Four, in every build including the free one. Indoor, ten classes: 81.9 % and 82.3 % held-out accuracy. Outdoor (LiDAR), aerial land cover: 86.2 %. Powerlines: 87 to 88 %. Streetscape, fifteen classes of street furniture: 31 of 32 traffic-sign objects found. Accuracy is per point, on data the model was not trained on.
How large a point cloud can it handle?
The limit is the disk, not the memory. The index is built out of core with bounded RAM and memory-mapped back: 43.5 million points open in about 1.2 s, and a 127.8 million point, 7.9 GB E57 in about 22 s. Very large clouds are classified in tiles automatically.
More in the FAQ.
Related guides
confidence column, and split a powerline corridor into individual trees, spans and poles.
Understand the scene
Decompose into primitives
Turn a scene of millions of points into a few hundred pieces a person can actually review (planes by size, cylinders, linear runs and clusters), each point carrying its component and its primitive_kind.
Build deliverables
Meshing (Poisson, clean mesh, segment meshes)
Three ways to a surface - screened Poisson for a smooth watertight shape, the clean mesher for a closed solid or an open façade with a texture baked from the cloud, and segment meshes for clean planar panels - each with an audit that says what the mesh is.