Think of messy, hand-drawn floor plans on rough construction drawings. This system is like an AI draftsman that can read those messy sketches, understand what is a wall, a room, a doorway, etc., and then convert them into a clean, structured digital floorplan automatically.
Manual digitization and cleanup of rough architectural floorplan drawings is slow, error-prone, and expensive. This research automates the segmentation and structuring of vectorized roughcast floorplans so walls, rooms, and other components are separated and labeled for downstream CAD/BIM workflows.
Domain-specific model architecture (two-stream graph attention over vectorized floorplans) and any curated training dataset of roughcast architectural plans form the core defensibility; performance depends heavily on proprietary labeled floorplan data and integration into CAD/BIM toolchains.
Open Source (Llama/Mistral)
Unknown
High (Custom Models/Infra)
Training and inference cost on large graph-structured representations of complex floorplans; collecting and labeling enough vectorized roughcast plans for robust generalization.
Early Adopters
Unlike generic image segmentation, this approach operates on vectorized floorplans and uses a two-stream graph attention network, better matching the underlying geometric and topological structure of architectural drawings and improving segmentation accuracy on noisy, roughcast inputs.