Floor Plan Intelligence
The goal: upload any floor plan and have a model find the walls, doors and windows, then place the proposal’s devices automatically. Step one is a labelled dataset; the model is still being built.
- Role
- In progress: labelling tool built, 4,035 images labelled, model still being built
- Status
- TRAINING
- Stack
- Python, Computer vision, Image annotation
The gap I spotted
Since joining AHA I have wanted a model that reads any floor plan and places the devices on it automatically.
The problem
Plans arrive as electrical, looping or architectural drawings, and no two are drawn the same way. I wanted a model that reads any of them, finds the walls, doors and windows, and places a proposal’s devices automatically: one less job for consultants, and something a customer can watch happen. My first detection attempts did not hold up because the plans are not standardised, so I went back to building a proper labelled dataset.
The system
- A keyboard-driven labelling tool that sorts floor-plan images into Architecture, Electrical, Looping or Trash, with undo, zoom and rotate.
- Annotating electrical plans for training computer vision models.
- 01
Raw floor plan
- 02
Annotation
- 03
Labelled dataset
- 04
Model training (in progress)
- 05
Plan understanding
- 06
Device and room logic
- 07
Proposal
Evaluation
- No accuracy figure yet. The model is still training, and a number from an unfinished model would mislead.
Impact
The labelling set holds 4,035 floor-plan images.
Limitations
- No model is good enough to publish yet, so there is no accuracy figure.
What I learned
- Dataset creation was a real engineering problem before any model: not every file that arrives is a usable plan, which is why Trash is a class.
AHA Smart Homes screens on this site are recreated with dummy data.
Want to talk through how I built this?
I'm happy to walk through the decisions, what failed and what I'd change.