About this project

SamGIS is an attempt to perform machine learning instance segmentation on geospatial data even without the use of dedicated graphics cards. The user interact on a leaflet webmap choosing areas to recognize and a backend responds with a geojson containing one or more recognized polygons within the initial webmap.

The backend perform machine learning inference using a Segment Anything model and ONNXRuntime as runtime.

Check about implementation details on my blog.

Self-hosted demo

You can visit my self-hosted demo here. Since this demo uses a python container backend hosted on my AWS account I keep it under authentication to prevent abuses. Here the OpenAPI swagger documentation.

HuggingFace Space demo

I added also a demo on this HuggingFace Space. It's possible to find here the direct space url and here the OpenAPI swagger documentation.