pointcloud2ifc -- Automated Scan-to-BIM Pipeline

Automated point cloud to IFC (BIM) conversion with CI/CD integration. The first open-source tool that offers automated IFC generation in CI -- drop a scan into your repo and get BIM output as a build artifact.

Pipeline

Point Cloud (PLY/PCD/LAS)
  |
  v
[1. Load & Downsample]  -- open3d voxel downsampling
  |
  v
[2. Segment]             -- DBSCAN clustering / RANSAC plane extraction
  |
  v
[3. Label]               -- Heuristic classification into BIMNet categories
  |
  v
[4. Build IFC]           -- Bounding-box geometry per segment (IfcExtrudedAreaSolid)
  |
  v
IFC File (.ifc) + JSON Report

CI/CD: Automated Scan-to-IFC in GitHub Actions

This is the key differentiator. No other tool offers automated IFC generation as a CI/CD pipeline.

Use from another repository (reusable workflow)

# .github/workflows/convert.yml in YOUR repo
name: Convert scans to IFC
on:
  push:
    paths: ["scans/**"]

jobs:
  scan2ifc:
    uses: MapIV/pointcloud2ifc/.github/workflows/scan2ifc.yml@main
    with:
      point_cloud_path: scans/building.ply
      method: ransac
      voxel_size: "0.05"

The IFC file and a JSON report are uploaded as build artifacts automatically.

Manual trigger (ad-hoc conversion)

Go to Actions > Scan-to-IFC Pipeline > Run workflow and specify:

• Point cloud file path
• Segmentation method (dbscan, ransac, or ml)
• Output path (optional)
• Voxel size (optional)

Segmentation methods

Method	Description
dbscan	DBSCAN clustering followed by heuristic labelling based on surface normals and geometry
ransac	Iterative RANSAC plane extraction; planes classified by normal orientation
ml	Pretrained PointNet segmentation (requires pip install 'pointcloud2ifc[ml]')

BIMNet 14-category labels

ID	Category	IFC Class
0	wall	IfcWall
1	floor	IfcSlab
2	ceiling	IfcCovering
3	door	IfcDoor
4	window	IfcWindow
5	column	IfcColumn
6	beam	IfcBeam
7	stair	IfcStairFlight
8	railing	IfcRailing
9	furniture	IfcFurnishingElement
10	slab	IfcSlab
11	curtain_wall	IfcCurtainWall
12	roof	IfcRoof
13	other	IfcBuildingElementProxy

Installation

pip install -e .

For development (includes pytest and ruff):

pip install -e ".[dev]"

For ML-based segmentation (PointNet backbone, requires PyTorch):

pip install -e ".[ml]"

For LAS/LAZ support:

pip install laspy[lazrs]

Usage

Convert a single point cloud to IFC

pointcloud2ifc convert input.ply -o output.ifc --method ransac --voxel-size 0.05

Supported input formats: PLY, PCD, LAS/LAZ.

Batch-convert a directory

pointcloud2ifc batch scans/ --output-dir output/ --method ransac --voxel-size 0.05

Processes all PLY, PCD, and LAS/LAZ files in the directory. Generates an IFC file and JSON report for each input, plus a batch_summary.json.

Evaluate against ground truth

pointcloud2ifc evaluate generated.ifc ground_truth.ifc --iou-threshold 0.5

Python API

from pointcloud2ifc.pipeline import Scan2IFCPipeline

pipeline = Scan2IFCPipeline()

# Single file
report = pipeline.run("scan.ply", "output.ifc", method="ransac")
print(f"Segments: {report.segments_found}, IFC elements: {report.ifc_elements_created}")

# Batch
reports = pipeline.run_batch("scans/", "output/", method="ransac")

Run tests

pytest tests/ -v

Architecture

src/pointcloud2ifc/
  __init__.py        # BIMNet category definitions
  cli.py             # Click CLI (convert, evaluate, batch)
  io.py              # Point cloud loading, IFC writing
  segmentation.py    # Semantic segmentation (DBSCAN, RANSAC, ML)
  pretrained.py      # Pretrained PointNet segmentation backend
  ifc_builder.py     # IFC model construction from segments
  pipeline.py        # End-to-end pipeline with batch processing and JSON reporting

License

MIT