Skip to content
Neurones 3D Download

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.

Where Process › Surface mesh (Poisson)… · Process › Mesh (clean)… · Process › Mesh Segments Tier Free + Pro CLI mesh-clean

Meshing is free. Writing the mesh as glTF / GLB or USD is Pro.

Part of Gaussian splatting software, Benchmarks, Gaussian splatting to mesh.

A clean mesh of the demo room, shaded, with the Topology line in the Properties pane reporting that it is not watertight and how many edges are non-manifold
The mesh says what it is: the Topology line reports holes and seams.

When to use it

Poisson for objects and smooth surfaces. Mesh (clean) for rooms, buildings and scans that need a closed solid, or tick Open scan for a façade scanned from one side. Mesh Segments for a lightweight model of the flat surfaces: walls, floors, table tops.

Step by step

  1. For Poisson, open Process › Surface mesh (Poisson)…, choose a Fidelity (Fast, Balanced, High or Extreme, depth 6 to 9, Balanced by default) and press Reconstruct. The window gives a time estimate as you choose.
  2. For the clean mesher, open Process › Mesh (clean)…. Leave Open scan (single-sided) off for a closed, watertight solid; tick it for a façade or wall scanned from one side. Choose Detail: Fast (about 200 k points) or Balanced (about 500 k) and press Mesh.
  3. For segment meshes, segment (or classify) first, then Process › Mesh Segments: one flat polygon per planar segment.
  4. Read the mesh in Properties › Mesh display: Shading (Shaded, Albedo, Normals, Flat), and the Topology line, which reports holes, the largest hole’s area, and seams.
  5. Write the mesh with Export… in Mesh display, or File › Export segment mesh… for segment meshes: .glb, or .usda for a USD stage.
  6. To go back from a mesh to points (for classification or export), use Convert to point cloud… in Mesh display or the Process menu.
The Mesh (clean) window: closed or open scan, detail Fast or Balanced, and the Mesh button
Closed by default; tick Open scan for a façade seen from one side.
The Convert to Point Cloud window: source line, a points slider set to 5 million, and the option to remove floaters and noise
Convert to point cloud: splats or a mesh become a real cloud the rest of the pipeline can work on.

What it was measured at

Measured

The clean mesher is closed, not edge-manifold, and says so: measured at 501 non-manifold edges on a 20,000-point synthetic sphere and 2,431 on a real indoor scan. Its texture is baked from the cloud into an 8K atlas.

From the command line

n3d-render-cli mesh-clean --input room.las --out room.glb             # closed, textured
n3d-render-cli mesh-clean --input facade.las --out facade.glb --open   # single-sided
n3d-render-cli mesh-clean --input part.las --out part.glb --poisson --depth 8
n3d-render-cli seg-mesh --input room.las --out panels.glb

Limits, stated

  • Poisson uses at most 800,000 points, a uniform subset of larger clouds.
  • No mesh decimation and no manifold repair are built; OBJ is not written.
  • Closed mode seals unobserved space by flood fill. A room with an open door is closed across the door; use Open scan when that is wrong.

Related questions

Is the clean mesh manifold?

It is closed, not edge-manifold, and N3D says so. Dual contouring puts one vertex per cell, and where two sheets of surface pass through one cell that vertex serves both: measured at 501 non-manifold edges on a 20,000-point sphere and 2,431 on a real indoor scan. The mesh audit reports boundary edges, non-manifold edges, shells and genus.

Can it simplify a mesh or write OBJ?

No. There is no mesh decimation, and OBJ is read-only. Meshes are written as glTF / GLB.

Can it repair a mesh into a manifold?

No. Manifold repair is not built. The mesh audit measures and reports the defect instead of claiming it away.

More in the FAQ.

About

Why Neurones 3D exists.

Standing inside a 43.5 million point indoor laser scan rendered by Neurones 3D

I have spent years working with point clouds, and for most of them the software got in the way. I wanted one tool I would actually reach for every day: fast enough to enjoy, honest about what it does, and calm to use.

Neurones 3D started in Python, the way most of my research does. It worked, but it was heavy, and heavy tools quietly discourage you from opening them. So I rebuilt everything in Rust, from the file readers to the renderer, with one rule: stay frugal. A single small file, no cloud, no telemetry, billions of points on a laptop that is ten years old. In an age where every tool reaches for a data center, I wanted the opposite.

The goal is simple to say and hard to earn. One package that carries you from capture to understanding to deliverable, without exporting to five other programs along the way. Load a scan, a splat, a building, a phone video. See it, classify it, measure it, and send out something real. Together, in one place.

This is the tool I always wanted. I hope it becomes yours too.

Dr. Florent Poux 3D Geodata Academy

Get Neurones 3D

Free tier, no account, one 15 MB file. We send the download link to your inbox.

Check your inbox

We sent a confirmation to your address. Click the link inside and the download follows right after. The link stays valid for 72 hours.

If Windows shows "Windows protected your PC", that is SmartScreen meeting a build that is not code-signed yet, not a virus: choose More info, then Run anyway. Check the file's fingerprint first if you like.

Double opt-in, no spam. Your address never leaves 3D Geodata Academy. Privacy.

Cookie settings

Choose what this site may store in your browser. Strictly necessary items are always on, because the feature you asked for does not work without them. Everything else is off until you switch it on. Your choice is kept in this browser for six months.

Strictly necessaryAlways on

Needed for a feature you use. Never used to track you.

NameKindWhat it is forKept
n3d_boardCookie, HttpOnlyA random voter token set by the board (roadmap), so one browser counts as one vote.12 months
n3d-feedback-draftLocal storageA draft of the feedback form, so a refresh does not lose your message. It never leaves your browser and is cleared when you send.Until sent
n3d_consentLocal storageThis choice.6 months
Analytics

None in use today. If a measurement tool is ever added, it stays off until you allow it here.

Media

None. Nothing on this site embeds a player or media from another company.