Code from A Comparative Evaluation of Deep Learning Techniques for Photovoltaic Panel Detection from Aerial Images (IEEE Access).

Paper available at: https://ieeexplore.ieee.org/document/10122915

• Clone the repository
• Create a new environment, e.g. python3 -m venv .venv
• Install requirements, pip install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html

Everything is launched through the run.py file, using a combination of click and pydantic. For more information about commands and their options, use python run.py [command] --help.

To launch a simple training, run python run.py train-segmenter --data-folder=... [args]. This will generate an output folder with a specific name, if provided, or simply the current timestamp. Inside each experiment directory, you'll find model checkpoints, output and tensorboard logs and the launch config.

Check the launch script for an example of how to launch an experiment.

To test the same experiment, launch python run.py test-segmenter --data-folder=... --output-folder=NAME_OF_THE_EXPERIMENT [other arguments, e.g. encoder type] The name is crucial, so that the task can find the right directory.

Check the launch script for an example of how to launch a test on the test set. Use instead the prediction script to generate predictions on a series of large rasters.