AI2BIM prototypes an end-to-end pipeline that converts images and generated meshes into editable IFC (conceptual LOD-200) massing. It combines controlled image synthesis, 3D reconstruction, point-cloud segmentation and parametric extraction to produce clean walls, slabs, roofs and voids suitable for early design exploration

• Convert sketches/NeRF/meshes to editable conceptual BIM (LOD-200)
• Prioritise spatial legibility, massing and openings over finish detail
• Enable rapid iteration and clean handoff to detailed design (not as-built)

Year: 2025   |   Type: Thesis Research
Methods: segmentation, monocular depth, SfM, graph optimisation, IFC
Tools: PyTorch, OpenCV, COLMAP, Blender, Rhino/Grasshopper, Rhino.Inside.Revit, IfcOpenShell