To design and develop an intelligent system that identifies hospitals and healthcare facilities from satellite images.
- Collecting the names of major cities in Maharashtra.
- Using Google Maps API to get the addresses of hospitals in these cities, including their coordinates.
- Using Mapbox API to get the aerial images for the collected coordinates of hospital buildings.
- Use CVAT Online to annotate these images with the help of Google Earth Pro.
- Convert the annotations in XML files to binary masks.
- Split the data into Urban and Rural areas, and then into train and test sets.
- Split the train data into train and validation data and train the UNet model.
- Use the model weights to make predictions.
- input: The address (and coordinates) CSV's from Google Maps API, the image files from Mapbox and the respective annotation XML files from CVAT.
- models: The final model weights.
- notebooks: Jupyter notebooks used for data visualization and other purposes.
- src: All the python scripts associated with the project.
- inference: Contains the geotiff infra maps, tile numbers containing infrastructure, images used for inference & predictions in .npz and geojson format.
-
How to clone the repo to start developing?
A:git clone https://github.com/Gokul-S-Kumar/Hospital_detection_from_aerial_images -
How to set-up the dev environment?
A: Use the "requirements.txt" file to clone the environment in which the model was developed. -
How to build on the existing work?
A: Use the model weights available here as pre-trained weights and develop new strategies to further improve the predictions.