Yours, and offline
Your data never leaves your machine. No account, no telemetry, and a licence key checked on the computer itself.
About Neurones 3D
From one point at a time to billions.
Neurones 3D began with a surveyor's question, asked alone in the office the night before a deadline: couldn't this be automated? This is the answer, built as one engine you can hold in a single file.
The question
It started with mud-caked boots.
Before the research and the code, Florent Poux was a land surveyor: out on site with the instrument, then back at the desk translating the real world into digital plans, one point at a time.
One evening stayed with him. A file of 2,587 surveyed points sat on the screen, a topographic map was due the next morning, and the work ahead was the same work as the day before. That is when the question arrived, and it has not left since.
Every feature in Neurones 3D is a version of it. Not can we look at the scan, but can the scan turn itself into the deliverable: the plan, the section, the terrain model, the volume, the report.
Going through these monotonous data points, I had my spark moment: "Couldn't this be automated?" That question, fuelled by a late-night cup of coffee, changed everything.
Who makes it
One researcher, fifteen years, one problem.
Florent Poux has spent fifteen years on a single problem: making reality capture answer for itself. First on site with a survey tripod, then in the lab, where his Ph.D. at the University of Liège gave the problem a name, then as a professor, and now at the academy he directs.
"I did not set out to build software," he says. "I set out to stop wasting good data." Neurones 3D is where that research stops being a paper and becomes a program you can run.
- Ph.D. 2019University of Liège, on the Smart Point Cloud; formerly adjunct professor in 3D geodata
- 60+ paperspeer-reviewed, with 1,700+ Google Scholar citations
- ISPRS awardthe Jack Dangermond Award, 2019
- O'Reillyauthor of 3D Data Science with Python (2025)
From the lab to everyone
State-of-the-art research, in a program anyone can open.
Research on point clouds usually stays where it was made: in papers, in lab code that needs its author to run it, on machines most offices do not have. Neurones 3D takes the Smart Point Cloud, the idea at the centre of that Ph.D., and puts it in one offline download that runs on an ordinary laptop.
A smart point cloud is one where every point knows not only where it is, but what it is, what it relates to, and how the next process can use it.
Where it is
Every point, fast
Billions of points streamed from disk out of core: LAS, LAZ, E57, PLY and Gaussian splats, at interactive frame rates.
What it is
Semantics
Segmentation, primitives and four trained classifiers: indoor, aerial LiDAR, streetscape and powerlines.
What it relates to
Structure
Scene graphs of rooms and objects, their adjacency, and the spaces a building is made of.
How it is used
Deliverables
Floor plans, sections, terrain rasters, change maps, volumes and reports, written by the engine itself.
The road
A surveyor, a scientist, a teacher, and then a toolmaker.
The learning curve, in his words, was steeper than the Pyrenees climbs he rides for fun. The path went from the field to the lab, into industry, and back out to the people who do the work.
01
Land surveyor
The field, the instrument, and plans drawn one point at a time. Where the question was asked.
02
Ph.D., postdoc, professor
From a blue-collar job to a Ph.D. at the University of Liège on the Smart Point Cloud, then years of research and a professorship.
03
Industry
Helping companies put 3D data to work, from optimising factory layouts to designing medical implants.
04
3D Geodata Academy
Director of the academy and author of 3D Data Science with Python (O'Reilly). Teaching professionals to ask the question for themselves.
The first public release
Open everything: LAS, LAZ and COPC, E57 with its 360° panoramas, and Gaussian splats, rendered out of core. Free for Windows.
Comparison and registration
Deviation in seconds, clouds aligned automatically, USD models written, and a point cloud you can send someone.
Drawings you would sign
Dimensioned floor plans with sections and elevations, scene graphs of rooms, M3C2 change detection and USD import.
Why an engine
Heavy tools quietly discourage you from opening them.
Neurones 3D started in Python, the way most of the research did. It worked, but it was heavy. So it was rebuilt from nothing in Rust, from the file readers to the renderer, under a single rule: stay frugal.
"In an age where every tool reaches for a data centre, I wanted the opposite."
One file. No installer, no account, no cloud, no telemetry. Billions of points on ordinary hardware, because the index streams from disk instead of filling memory. And one package that carries you from capture to understanding to deliverable, without exporting to five other programs along the way.
What it stands for
Six promises, and how to check each one.
Measured, not promised
Every speed and accuracy figure on this site was measured on named data, and the reports say how.
Frugal by design
Pure Rust, one executable, and an out-of-core index, so billions of points open on a ten-year-old laptop.
Deliverables, not views
Plans, sections, terrain rasters, change maps, meshes and reports: the thing you were going to be paid for.
Limits, stated
Every guide ends with what the process does not do, and every figure keeps its level of detection beside it.
Taught, not hidden
The science behind each process is written down in the book and taught at the academy, not locked in a black box.
By the numbers
An engine you can measure.
1.2 sto open a 43.5 million point scan
60 fpswhile it renders, out of core
140klines of Rust, in seven crates
1,014automated tests, all passing
4trained classifiers, in every build
0accounts, uploads and telemetry calls
Measured on the engine's own benchmark harness and test suite, September 2026.
Made by the engine
Every render on this site comes out of Neurones 3D.
The science behind it
Use the tool. Or learn to build one.
The book
3D Data Science with Python
Published by O'Reilly. From loading a first point cloud to segmentation, registration, meshing and 3D deep learning: the ideas that Neurones 3D puts into a single program.
Read about the book → 3D Geodata AcademyCourses for 3D professionals
The academy Florent Poux directs teaches point cloud processing, 3D machine learning and geodata workflows to surveyors, engineers and researchers.
Visit learngeodata.eu →This is the tool I always wanted. I hope it becomes yours too.
Free for Windows, forever. Open your own scan, or try one of the demo scenes built into the app.
Dr. Florent PouxDirector, 3D Geodata Academy · author of 3D Data Science with Python