Various scripts have to be executed in sequence, they are numbered from 0 to 4.
Adds a raster of the polygon for each sample of the tfrecords for U-Net training.
The 1_train.py script performs the training of the model on the train dataset and validates on the val dataset.
Visualize training using TensorBoard (you might have to modify the --logdir argument according to your directory structure):
python -m tensorboard.main --logdir=~/polycnn/code/unet_and_vectorization/runs
Some parameters at the beginning of the script can be changed.
Compute polygon predictions on the test set using the trained model + vectorization and saves the results (used later on to measure accuracy performance).