How ArchLab Converts Sketches into Dimension-Accurate Floor Plans
A three-step pipeline powered by deterministic computer vision — no generative hallucination, no approximate guesses. Every dimension is derived from real geometry.
Upload Your Sketch, Scan, or Photograph
ArchLab accepts virtually any architectural input: hand-drawn sketches on napkins, scanned floor plans from paper archives, site photographs taken on a smartphone, or even rough PDFs from clients.
Our intake engine normalises orientation, corrects lens distortion from photos, and identifies architectural primitives — walls, doors, windows, and structural elements — before any measurement begins.
- Accepts JPEG, PNG, PDF, TIFF, HEIC, and WebP
- Automatic orientation and lens correction
- Primitive detection: walls, openings, columns, stairs
- Works with incomplete or partial sketches
Deterministic Dimension Extraction
This is where ArchLab fundamentally differs from generative AI tools. Instead of predicting what a room might measure, our pipeline extracts real dimensions using geometric constraint solving.
Walls are straightened to their true alignment. Angles snap to structurally plausible values (90-degree, 45-degree, or as-built). Every measurement is derived from the spatial relationship between detected primitives — not hallucinated by a language model.
- Geometric constraint solver, not a generative model
- Wall straightening and angle snapping
- Sub-centimetre accuracy from calibrated inputs
- Zero dimension hallucinations — if it cannot measure, it flags uncertainty
Export to CAD, 3D Render, or Both
Once dimensions are extracted, ArchLab generates production-ready output in the format your workflow demands. Editable DXF and DWG files open directly in Revit, AutoCAD, ArchiCAD, and other BIM software.
Need a client-ready visual? Our rendering engine produces photorealistic 3D views from the same extracted floor plan — no additional modelling required.
- DXF/DWG export compatible with Revit, AutoCAD, ArchiCAD
- Photorealistic 3D renders from floor plan data
- Layer-separated output for walls, doors, windows, dimensions
- Batch processing for multi-room or multi-floor projects
Why Deterministic AI Matters for Architecture
Generative AI hallucinates. In architecture, a hallucinated dimension means a wall that does not fit, a door that does not open, or a structural element in the wrong place. ArchLab takes a different approach.
No Hallucinations
Every measurement output is derived from geometric relationships in your input. If ArchLab cannot determine a dimension with confidence, it flags it rather than guessing.
Reproducible Results
Feed the same input twice, get the same output twice. Our deterministic pipeline eliminates the randomness inherent in generative models.
Professional-Grade Output
Layer-separated CAD files, proper annotation, and BIM-compatible formats. Output that fits into real architectural workflows, not demo-ware.