# Generating IFC Models from 3D Gaussian Splatting (3DGS)
> Learn how to bridge the semantic gap between 3D Gaussian Splatting and structured IFC building models using Python-based automated workflows for scan-to-BIM.

Tags: 3d gaussian splatting, ifc-model, bim, scan-to-bim, 3dgs, python, digital-twin, construction-ai
## Semantic Enrichment of 3D Gaussian Splatting for IFC Model Generation

*   **Objective:** To bridge the gap between photorealistic 3DGS reconstructions and semantically rich IFC building models for BIM workflows.
*   **The Problem:** 3DGS produces fast, high-quality geometry but lacks the structured alphanumeric data required for industry standards (IFC).

## Technical Background

*   **3DGS:** A novel view synthesis method that optimizes 3D Gaussian primitives. It achieves 100+ fps rendering at 1080p and trains in under 1 hour.
*   **IFC:** Open standard (ISO 16739) for BIM exchange. Requires geometric info, semantic classification, and spatial relationships.

## Developed Python Workflow & Tools

1.  **CC_cloud_patch.py:** Fills point cloud holes using convex hull and grid interpolation.
2.  **CC_3D_mesh.py:** Converts point clouds to extruded prism meshes using PCA and bounding boxes.
3.  **mesh2ifc.py:** Converts meshes to IFC with auto-classification for walls, slabs, windows, and doors.

## Key Results

*   **Data Efficiency:** Achieved a data reduction ratio of ~12,000:1 (2.6 million points reduced to 15 IFC elements comprising 230 vertices).
*   **Validation:** Successfully generated an IFC model with correct spatial hierarchy (IfcProject -> IfcSite -> IfcBuilding -> IfcBuildingStorey) validated in BIMcollab Zoom.
*   **Limitations:** Current workflow assumes planar elements and uniform thickness; manual segmentation remains the most time-consuming step (approx. 80% of processing time).
---
This presentation was created with [Bobr AI](https://bobr.ai) — an AI presentation generator.