BIM-to-BIM Matching

Compute per-element 3D-IoU and 3D-Compactness metrics between two IFC files (ground truth vs. predicted). Preserves full IFC metadata throughout the process.

Features

• Load IFC files preserving GlobalId, Name, IfcType, attributes, and property sets
• Align predicted model to ground truth (ICP or centroid-based)
• Compute pairwise IoU matrix between elements
• Calculate per-element metrics for both GT and predicted models
• Export results to JSON and CSV formats

Installation

pip install -r requirements.txt

Requirements

• Python 3.8+
• numpy
• open3d >= 0.18.0
• ifcopenshell >= 0.7.0
• trimesh

Usage

python main.py --gt <ground_truth.ifc> --pred <predicted.ifc> [options]

Arguments

Argument	Description	Default
--gt	Path to ground-truth IFC file (required)	-
--pred	Path to predicted/reconstructed IFC file (required)	-
--align	Alignment method: icp, centroid, or none	icp
--backend	CSG backend: auto, exact, or voxel	auto
--voxel-size	Voxel pitch in meters for fallback operations	0.03
--epsilon	IoU threshold to consider a correspondence	0.05
--save-json	Path to save JSON report	-
--save-csv-prefix	Prefix for CSV output files	-
--ifc-classes	Filter by IFC classes (e.g., IfcWall IfcDoor)	all

Example

python main.py \
    --gt ground_truth.ifc \
    --pred reconstructed.ifc \
    --align icp \
    --backend auto \
    --voxel-size 0.05 \
    --epsilon 0.1 \
    --save-json metrics.json \
    --save-csv-prefix out/metrics \
    --ifc-classes IfcWall IfcSlab

Output Files

File	Description
metrics.json	Complete metrics report in JSON format
<prefix>_per_gt.csv	Per-element metrics for ground truth elements
<prefix>_per_pred.csv	Per-element metrics for predicted elements
<prefix>_edges.csv	Pairwise IoU values between matched elements

Helper Scripts

The helpers/ directory contains utility scripts:

• move_storey.py - Edit IFC storey structure (rename, remove, reassign elements)
• rotate_180.py - Rotate all storeys and elements by 180° around Z-axis
• print_spaces.py - List and analyze IfcSpace boundary relationships

Citation

This implementation is inspired by the evaluation metrics proposed in the following paper:

Liu, Y., Huang, H., Gao, G., Ke, Z., Li, S., & Gu, M. (2025). Dataset and benchmark for as-built BIM reconstruction from real-world point cloud. Automation in Construction, 173, 106096. https://doi.org/10.1016/j.autcon.2025.106096

@article{liu2025dataset,
  title={Dataset and benchmark for as-built BIM reconstruction from real-world point cloud},
  author={Liu, Yudong and Huang, Han and Gao, Ge and Ke, Ziyi and Li, Shengtao and Gu, Ming},
  journal={Automation in Construction},
  volume={173},
  pages={106096},
  year={2025},
  publisher={Elsevier},
  doi={10.1016/j.autcon.2025.106096}
}

License

See LICENSE file for details.