This project provides a small FastAPI service that detects changes between two images. It uses the Segment Any Change model to produce an overlay mask and can optionally describe the differences with a vision language model.

• REST API built with FastAPI
• Change detection using pre-trained Segment Any Change weights
• Optional description of changes via Gemini or Ollama
• Dockerfile and docker-compose configuration for GPU inference

Create a Python virtual environment and install the dependencies:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Some packages (e.g. PyTorch models) can take a while to install. Alternatively, you can use the provided Dockerfile or docker-compose.yaml for a containerised setup.

To launch the API locally after installing the requirements:

cd app
uvicorn main:app --host 0.0.0.0 --port 7866 --reload

Open your browser at http://localhost:7866 to access the demo page.

If you prefer Docker, build and run the image:

docker-compose up --build

Send a POST request to /detect with two image files (ref_img and test_img). The response contains logs, the percentage of changed pixels and a link to the mask overlay image.

Example using curl:

curl -F "[email protected]" -F "[email protected]" http://localhost:7866/detect

This project is released under the MIT License. See LICENSE for details.