Point cloud processing
Point cloud processing software for Windows
Neurones 3D (N3D) is point cloud processing software that runs on one Windows PC, offline, from a single executable. It opens the files scanners write, draws billions of points out of core, and adds point cloud intelligence: classification, segmentation, rooms and objects, then the deliverables built from them, each with the number it was measured at. It is built on the Smart Point Cloud, the research of Dr. Florent Poux’s Ph.D. at the University of Liège: a point cloud whose points know not only where they are, but what they are, what they relate to, and how the next process can use them.
By Dr. Florent Poux, 3D Geodata Academy · Updated
At a glance
Short statements, each one checkable. Where there is a number, it was measured, and the guides say on what.
- N3D opens LAS, LAZ, COPC, E57, PLY, SPZ, PCD, NPY, OBJ, STL, glTF, IFC, STEP, USD and ASCII point files.
- A 43.5 million point LAS file opens in about 1.2 s and renders at 60 FPS; a 127.8 million point E57 opens in about 22 s.
- The point index is built out of core with bounded memory, so a cloud larger than the machine’s RAM still opens.
- Four trained classifiers ship in every build: indoor (82 % held-out), outdoor aerial LiDAR (86.2 %), powerline (87 to 88 %) and streetscape.
- The ground filter scores F1 90.3 % against IGN LiDAR HD’s own ground labels, with no training data.
- It builds floor plans with sections and elevations, terrain products, M3C2 change maps, meshes and HTML reports with provenance.
- It is one Windows executable with no installer, no account and no telemetry; the licence is checked offline.
- There is a command line,
n3d-render-cli, with more than thirty subcommands that load through the same cache as the app.
What it does, in the order you use it
Load and look
Drop a file on the window. Scans are indexed once into a local cache and memory-mapped back, so the second open takes about a second. Eye-dome lighting, a continuous-surface mode, surfels and a voxel view are one click apart, and three quality presets fit the drawing to the GPU.
Understand
Unsupervised segmentation splits the scene into surfaces; the decomposition turns 43.5 million points into 1,501 reviewable pieces in 5.8 s; four classifiers label them, each writing a per-point confidence; a labelling grid lets a person correct what the model got wrong.
Build and deliver
From the same understanding: a scene graph with rooms and openings, a dimensioned floor plan with sections and elevations, terrain rasters on one grid, M3C2 change detection with a level of detection, clean meshes, and an HTML report whose provenance block lets someone else re-derive every number.
Questions
What is point cloud processing software?
Software that reads the point clouds produced by laser scanners, LiDAR and photogrammetry, and turns them into information: classes, objects, measurements, drawings and models. N3D does the reading, the understanding and the deliverables in one application.
What is a smart point cloud?
A point cloud that carries meaning, not only coordinates: each point knows what it is (its class and segment), what it relates to (the object, room or surface it belongs to) and how a downstream process can use it. The idea comes from Dr. Florent Poux’s Ph.D. research at the University of Liège (2019); N3D is the software built on it.
Does N3D need the cloud or an internet connection?
No. Everything runs on your PC. There is no upload, no account and no telemetry, and the licence key is checked on the machine.
How large a point cloud can it open?
The limit is the disk, not the memory: the index is built out of core and memory-mapped. Measured: 43.5 million points in about 1.2 s, and a 127.8 million point, 7.9 GB E57 in about 22 s.
Is there a free version?
Yes. The free tier views, imports and runs every process in the viewer, including the four classifiers. The Desktop edition, bought once and yours to keep, unlocks writing out what the program computed.
What does it not do?
It does not write DWG, decimate meshes, repair meshes into manifolds, check SLAM trajectories, or run on macOS or Linux yet. The FAQ lists each, with the reason.
The guides behind it
Each guide gives the menu path, the steps, the command-line equivalent and the limits.
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
Floor plans, sections and elevations
One command from raw points to a drawing set - walls between their measured faces, doors with their swing, windows with their glass, dimension chains, room areas and levels, a legend, a scale bar and a north arrow - plus sections and elevations placed where they show the most.
Build deliverables
Terrain products (DTM, DSM, CHM, contours)
Turn one ground fit into the raster set a client actually receives (DTM, DSM, canopy height, slope, aspect, hillshade, contours, trees and sunlight), all on one grid so they cannot disagree, with a README that says what is wrong with them.
Compare and register
Compare and detect change (M3C2)
Compare a survey against a reference (another survey, a mesh or a BIM model) as a signed distance with conformance bands and a worst-first list of defects, or run M3C2 to say what was gained, what was lost, and what simply was not observed.
Deliver
Reports
One self-contained HTML document from what the session measured - the as-built review with its per-zone punch list, the volume and how much of it rests on the classification, the measurements you kept - and a provenance block that lets someone else re-derive every number.