Understand the scene
Classify a point cloud (four models)
Label every point with one of four trained models (Indoor, Outdoor aerial LiDAR, Powerlines, Streetscape), read how sure the model was from a per-point confidence column, and split a powerline corridor into individual trees, spans and poles.
Classifying is free, and the trained models ship in every build. Exporting the class and confidence columns is Pro.
Part of Point cloud processing software, Point cloud classification, LiDAR terrain (DTM, DSM).

When to use it
Whenever the question is what rather than where: how much of this scan is floor, which points are vegetation near the conductors, where the traffic signs are. Pick the model that matches the capture: an indoor terrestrial or handheld scan, an aerial LiDAR tile, a utility corridor, or a mobile-mapping street.
Step by step
- Open Process › AI classification… (under Classify), or the classifier card in Properties › AI Analysis. The models are listed down the left: Indoor (the default), Powerlines, Outdoor (LiDAR) or Streetscape (Dev). For the one picked, the window shows what it is for in one line and the classes it will paint as colour chips.
- Press Run classification. Running a model replaces the current classes. If the scene has not been segmented yet, N3D segments it first (
Segmenting, then classifying…). - When it finishes, the Classes point view comes on. The legend lists each class present with its point count and a box to show or hide it.
- Colour by confidence (Display › Colour by) to see where the model hesitated: it is the share of the forest’s trees that agreed on each point’s label.
- On a powerline corridor, press Instances to separate each tree, conductor span and pole into its own object (an
instancecolumn). - To correct labels, open Process › Label Components…: the grid switches to the model’s vocabulary. See label components.




What it was measured at
Held-out, per point: Indoor (ten classes) 81.9 % and 82.3 %; Outdoor (LiDAR) against IGN LiDAR HD 86.2 %; Powerlines 87 to 88 %; Streetscape found 31 of 32 traffic-sign objects.
The four models
| Model | Classes |
|---|---|
| Indoor | ceiling, floor, wall, structure, window, door, table, seat, storage, clutter |
| Outdoor (LiDAR) | ground, low vegetation, medium vegetation, high vegetation, building, other |
| Powerlines | ground, vegetation, building, powerline, pole or structure, other |
| Streetscape (Dev) | others, bushes, trees, buildings, cars, bridges, support columns, fences, traffic signs, other signs, traffic lights, poles, wires, forest, mailboxes |
Each is a random forest over segment, point and neighbourhood features, run on the CPU, on your machine. Nothing is uploaded and no GPU is needed for it.
Limits, stated
- Accuracy is per point and on data the model was not trained on. A capture unlike the training data (a different scanner density, an unusual building) will score lower; the
confidencecolumn is how you find where. - Instances is offered for the Powerlines model only, and not after a tiled pass, because it needs every point resident at once.
- Streetscape is a development preview.
- N3D does not certify its own classification. Review, a human override and an exportable baseline are how the result becomes something to sign.
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 do I know how sure the classifier was?
Each classification writes a confidence column: the share of the random forest’s trees that agreed. Colour by it to see where the model hesitated, threshold on it, and review those parts in Label Components…. N3D does not certify its own output; confidence, a human override and a baseline you can export are what it offers instead.
Will the classifier certify its own result?
No, deliberately. N3D reports a per-point confidence, lets a person review and correct every component, and keeps the model’s baseline beside the human labels. That is what can be signed; a model marking its own work is not.
Does classification need a powerful GPU?
No. The classifiers, segmentation, decomposition and the ground filter run on the CPU, on your machine. The GPU draws the scene.
More in the FAQ.
Related guides
segment column for every point.
Understand the scene
Ground filter
Drop a simulated cloth on the upside-down cloud to fit the ground, label every point ground or off-ground, and measure each point’s height above that ground, with no training data and parameters you can adjust live.
Understand the scene
Transfer a column from a layer
Work on a light subsample, then project the answer (a classification, a segment id, a height) onto the full-resolution cloud, and leave a point with no source in range empty rather than guessing.