A web-based tool for matching and labeling building footprints from two datasets. Runs locally as a Flask app, displaying Folium-generated maps of potential matching pairs.
pip install git+https://github.com/ai4up/eubucco-conflation@mainStep 1: Create a dataset of potential building pairs from two datasets:
gmatch create-labeling-dataset dataset1.parquet dataset2.parquetStep 2: Start browser-based labeling of building pairs:
gmatch labelCreate a dataset of potential matches of government buildings and Microsoft buildings for a small region in France that require manual labeling using the demo data in the repository. Include only buildings which overlap slightly (0-10%).
gmatch create-labeling-dataset \
--min-intersection=0.1 \ # Minimum relative overlap for new buildings to be included in labeling dataset [0,1)
--max-intersection=0.2 \ # Maximum relative overlap for new buildings to be included in labeling dataset (0,1]
data/demo-gov.parquet data/demo-microsoft.parquetThe resulting dataset is locally stored as candidate-pairs.pickle. To initiate the browser-based labeling, run:
gmatch labelPrerequisites
- Ensure a dataset of candidate pairs (
candidate-pairs.pickle) is present in thedatadirectory.- For production deployment, set a Flask session
SECRET_KEYenvironment variable.
Serve the dockerized Flask app with an Nginx proxy at http://localhost:80:
docker-compose upInstall dev dependencies using poetry:
poetry install --only devInstall git pre-commit hooks:
pre-commit installBuild from source:
poetry build
pip install dist/geo_matcher-*.whl