This is an AI system that looks at a rough, sketchy floorplan and automatically turns it into a clean, computer-understandable layout by figuring out where rooms and walls are. Think of it as a smart assistant that traces and labels an architect’s messy sketch into a neat digital plan.
Manual cleanup and digitization of rough floorplan sketches is slow, error‑prone, and expensive. This model automates the segmentation and labeling of roughcast (draft) floorplans in vector form, speeding up design workflows and enabling downstream automation (area calculations, code checks, BIM integration).
Research-grade model architecture (two-stream graph attention network) specialized for vector floorplans; potential access to curated training datasets of roughcast plans; integration into CAD/BIM workflows can create high switching costs once embedded.
Open Source (Llama/Mistral)
Unknown
High (Custom Models/Infra)
Training and inference on large, complex vector graphs (many nodes/edges) can be computationally heavy; performance may depend on availability of sizeable labeled floorplan datasets.
Early Adopters
Unlike generic image segmentation tools, this approach operates directly on vectorized roughcast floorplans and uses a two-stream graph attention network tailored to architectural drawings, which can yield more precise room/element boundaries and better integration with CAD/BIM data structures.