You have two options, you can either download the dataset or generate it yourself using the two notebooks dataset_extraction/positives.csv and dataset_extraction/positives.csv notebooks.
Fow downloading the data refer to the README of the main page of this repository.
With the mesogeos datacube downloaded (see mesogeos/README.md), you can generate the dataset using the two notebooks dataset-extraction/positives.ipynb and dataset-extraction/negatives.ipynb notebooks.
To reproduce the scripts you should also download the shapefile with the burned areas and the shapefile with the Ecoregions (available in the Google drive link provided in the main page of this repository) and add the paths to the empty path templates in the notebooks.
Install all the requirements from the requirements.txt file. Disclaimer: this environment file is not minimal, it contains all the packages used during the development of the project. Also, it is not guaranteed to work on all platforms. It has been tested on Ubuntu 20.04 with an NVIDIA RTX 3080 GPU.
pip install -r requirements.txt
To run the experiments that presented in the paper for the three models LSTM, Transformer and Gated Transformet Network, by running:
- train.py --config configs/config_lstm/config_train.json
- train.py --config configs/config_transformer/config_train.json
- train.py --config configs/config_gtn/config_train.json
Before running the experiments you should add the dataset path to the "dataset_root" in the config files.
Similarly for testing the results:
- test.py --config configs/config_lstm/config_test.json
- test.py --config configs/config_transformer/config_test.json
- test.py --config configs/config_gtn/config_test.json
Before running the tests you should add the dataset path to the "dataset_root" and the trained model path to the "model_path" in the config files