utkucanozturk/gather-project

Completing the road network of the Antananarivo (Madagascar) to stimulate sanitation serviceability. # Project Gather ## Summary Project with a purpose of completing the road network of the Antananarivo, Madagascar to stimulate sanitation serviceability. * Combining and assessing quality of existing digital maps, e.g. Sentinel-2, Google Maps and Mapbox, to improve the route-network to close sanitation data gap. * Implementation of a road segmentation model (U-net) on aerial imagery to detect routes that have not been digitized yet. ## Gathering Image Data You can run the get_images notebook to collect the images for the area of your choice. To be able to run the notebook, you need a shapefile of the area of interest which will be read as geodataframe. Geodataframe will be used to determine the locations that we will gather the images of. Geodataframe/shapefile can be in projection of any coordinate reference system. You can simply follow the instructions/notes inside the notebook to create your own dataset. Please note that notebook makes calls to the Mapbox Static Images API which has rate limits that cap the number of requests you can make against an endpoint. ## Modeling The model comparison part of the road detection project for Madagascar allows to train and compare the performance of different road segmentation models. The test set are currently images from the district 8 of the Antanarivo Province scraped from mapbox. The framework can easily be extended to further configurations. ### Usage The file model_comparison_demo.ipynb gives an example how to use the model comparison api and evaluate the results. The demo shows how to call the "perform_cv" function of the module module_comparison.py. Parameters are the IMAGE_FOLDER with the test_images, the MODEL_NAMES of the models to be evaluated, a model_getter function that determines how to make predictions from a model and a data loader function to load the data. The file test_model_comparison.py contains unit tests for the model comparison part. ### Setup The required libraries for the model compar …