Both the training set and the validation set are lists. Each entry in the list is a dictionary with keys “image” and “mask”. Both of these contain numpy arrays with image data.
The image data is captured from the European Space Agency’s (ESAs) Sentinel-2 satellite with a pixel size of 10 m x 10 m, meaning that each image corresponds to 2.56 km x 2.56 km. The three channels correspond to raw channel readings (values in the range from 0 to thousands) from the sensor in Blue-Green-Red order. The mask is a binary mask with 1s denoting pixels that contain buildings and 0s elsewhere.
The dataset should be placed into the same directory.
I used python 3.8 for this project:
conda create --name myenv python=3.8
To install the requirements, run in your virtual environment:
conda install --yes --file requirements.txt
To run the model on a dataset, please type:
python3 cli.py model_path dataset_path