Segmentation-Fault

for hackathon purpose

Prepare Data

Prerequisites

In order to smoothly reproduce the experiments, it is recommended to use Anaconda. You can find installation instruction here: https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html

Enviroment file is provided. Please use it to recreate the enviroment using following command:

conda env create -f req.yml

And activate the enviroment:

conda activate hackathon

You will also need to install appropriate version of gdal package.

For Ubuntu users, please follow the instruction: https://mothergeo-py.readthedocs.io/en/latest/development/how-to/gdal-ubuntu-pkg.html

Prepare Data

1. In order to prepare the data for the training, run:

   python src/data/crop_data.py -c src/configs/config.yml

   In the config file, you have to specify:

   • data_dir: path to directory shared by Varuna with the data
   • cropped_data_dir: target path, where the preprocessed data will be saved

   Please follow the provided example.

   If you would like to skip this step (as it is quite time cosuming and requires correctly installed gdal package), please download already prepared data from here: https://drive.google.com/file/d/1nQlKEAItAxCWUWoY-Y90lC6XHHNFwf3D/view?usp=sharing

Split Data

In order to split the data, run:

python src/data/split_data.py -c src/configs/config.yml

Please find the description of required parameters in provided config file exmaple from line above.

Train

You can configure training parameters from the config file. To recreate out best results model, please use the parameters provided in the example.

Run the training using following command:

python src/training/training_pipeline.py -c src/configs/config.yml

Evaluate

python src/evaluation/evaluate.py -c src/configs/evaluate_config.yml

Predict

python3 src/evaluation/predict.py -c src/configs/predict_config.yml