Skip to content

Mesh Segmentation

In one sentence: Automatically split a mesh into regions by classifying each face as plane, cylinder, sphere, or free-form using curvature analysis.

Overview

Workbench: Reverse Engineering
Menu: Reverse Engineering → Segmentation → Mesh Segmentation…
Shortcut: none

Runs a region-growing algorithm that classifies mesh faces by their principal curvatures, groups adjacent faces of the same type into segments, and creates one Mesh::Feature per segment inside a group named "Segments". The segments can then be fed individually to the Approximation tools.

Intuition

Curvature reveals what type of surface a face belongs to:

  • A face with near-zero curvature in both directions → flat plane
  • A face with curvature in one direction and zero in the other → cylinder or cone
  • A face with equal curvature in both directions → sphere

Mesh Segmentation measures these curvature signatures, assigns a label to each face, then groups connected faces with the same label into a segment. Think of it as automatically colour-coding your mesh by surface type.

When to use it

  • Your scan contains recognisable geometric primitives (flat faces, cylindrical bores, spherical fillets) and you want to fit analytic surfaces to them.
  • You want to divide a complex scan into manageable regions before applying B-spline fitting to each region.
  • You are automating a scan-to-CAD pipeline and need region labels for each primitive type.

When NOT to use it

  • Highly noisy meshes — noise inflates the apparent curvature of flat faces. Smooth the mesh first (Mesh workbench → Smoothing).
  • When you need manual control — use Manual Segmentation to paint regions interactively.
  • When the mesh is already split into components — use From Components instead.

Step-by-step

  1. Select a single Mesh::Feature.
  2. Choose Reverse Engineering → Segmentation → Mesh Segmentation….
  3. In the task panel, set:
  4. Smoothing steps — number of Laplacian smoothing passes before curvature estimation (start with 3–5).
  5. Plane / Cylinder / Sphere tolerance — curvature thresholds for each type (start with defaults and adjust if segments are wrong).
  6. Min region size — discard segments with fewer faces than this.
  7. Click OK. A "Segments" group is created with one Mesh::Feature per region.

Tip

Run Mesh workbench → Smoothing (1–3 iterations) on the input mesh before segmenting. Even a little smoothing dramatically improves classification accuracy on real-world scan data.

Parameters

Parameter Type Default Description
Smoothing steps Integer 3 Laplacian smoothing passes before curvature estimation
Plane tolerance Float 0.01 Max curvature magnitude for planar classification
Cylinder tolerance Float 0.01 Curvature asymmetry tolerance for cylinder classification
Sphere tolerance Float 0.01 Deviation from equal-curvature for sphere classification
Min region size Integer 100 Segments with fewer faces are discarded

Common mistakes and pitfalls

Too many tiny segments

Cause: Surface noise causes smooth faces to be split into many small patches with varying curvature labels.
Fix: Increase smoothing steps and Min region size. Apply Mesh workbench smoothing before segmenting.

Cylinders classified as free-form

Cause: Cylinder tolerance is too tight for the curvature noise level in the mesh.
Fix: Increase Cylinder tolerance, or use Manual Segmentation to paint cylinder regions.

Result has many unlabelled 'free-form' segments

Cause: The mesh has blending fillets, transitions, or complex surfaces that don't match any primitive type.
Fix: This is expected for complex geometry. Apply Approximate B-Spline Surface to the free-form segments.

Python API

# Mesh Segmentation is command-driven; trigger via GUI command:
import FreeCADGui as Gui
Gui.doCommand("import ReverseEngineering")
Gui.doCommand("Gui.runCommand('Reen_Segmentation', 0)")

See also