Click any feature on a georeferenced raster and SAM2 segments it instantly. Refine with positive/negative prompts, then accept to create a vector polygon. All processing runs locally — by LAD, Sapienza University of Rome.
AITracer by LAD brings AI-assisted raster digitization to QGIS 4.
Activate the tool, click any feature in a georeferenced raster —
aerial photo, drone orthophoto, scanned plan — and the SAM2-tiny
model segments it instantly as a polygon preview. Refine the result
with positive or negative point prompts, adjust Douglas-Peucker
simplification, then accept to add the polygon to any vector layer.
HOW IT WORKS: AITracer captures a screenshot of the current QGIS
canvas and sends it to SAM2 — it does not read the original raster
file. This keeps processing fast and memory-light on modest hardware.
The output resolution depends on your zoom level, not the raster
resolution. Features that extend beyond the canvas edge will be
partially traced. The canvas is locked during a session to keep
pixel-to-map coordinates consistent.
First-time setup automatically downloads a standalone Python runtime
and the SAM2-tiny model weights (~160 MB). An internet connection is
required for this one-time step only. All subsequent processing runs
entirely on your machine; no data is ever sent to external servers.
Developed by LAD — Laboratorio di Archeologia Digitale,
Sapienza University of Rome.
Plugin Tags