Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery

Pratinav Seth (#), Michelle Lin (#), Brefo Dwamena Yaw, Jade Boutot, Mary Kang & David Rolnick.

(# -Denotes co-first authorship.)

Dataset Access

The dataset is publicly available for academic research. You can access it directly from Zenodo: Alberta Wells Dataset on Zenodo. Given the large size, if you want to install using zenodo downloader then :

pip install zenodo_get
zenodo_get 13743323

(NOTE : If you want to run with the code in this repository dont download at this point follow instructions in next section)

Setup Project

Clone the Github Repo

git clone https://github.com/RolnickLab/Alberta_Wells_Dataset.git
cd Alberta_Wells_Dataset
export AWD_CODEBASE=$(pwd)

Download the Dataset

Create a setup conda enviroment

conda create --name awd_setup python=3.10
conda activate awd_setup

Download from Zenodo (HDF5 File)

cd $AWD_CODEBASE
mkdir downloads
cd downloads
pip install zenodo_get
zenodo_get 13743323
cat train_set.tar.gz.part_* > train_set.tar.gz
tar -xzvf eval_set.tar.gz
tar -xzvf test_set.tar.gz
tar -xzvf train_set.tar.gz
cd ..

Split HDF5 File into Single Instances for Faster Processing

python setup/file_handling_cc_train.py $PWD
python setup/file_handling_cc_test.py $PWD
python setup/file_handling_cc_eval.py $PWD

Running Experiments

Setup Experimets Conda Enviroment

cd $AWD_CODEBASE
conda create --name awd python=3.11.7
conda activate awd
cd setup
pip install -r requirements.txt
cd ..

Training

TORCH_USE_CUDA_DSA=1 python main.py --config=configs/<directory_location>/<filename>.json --CUR_DIR=$AWD_CODEBASE --SEED=333

To continue training from a checkpoint, make the following change in the config.json file :

    "checkpoint_file": "False",

to

    "checkpoint_file": <relative location of checkpoint in codebase>",

Inference

To run inference using a specific checkpoint after training, or to continue training from a checkpoint, make the following change in the config.json file :

  "max_epoch": 50,
  "checkpoint_file": "False",
  "data_mode": "training",

to

  "max_epoch": -1,
  "checkpoint_file": <relative location of checkpoint in codebase>",
  "data_mode": "inference_valid_test",

Contributions

If you find the dataset valuable for your research, please cite our work using the following reference:

@misc{seth2024albertawellsdatasetpinpointing,
      title={Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery},
      author={Pratinav Seth and Michelle Lin and Brefo Dwamena Yaw and Jade Boutot and Mary Kang and David Rolnick},
      year={2024},
      eprint={2410.09032},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.09032},
}