search

nannau / DoWnGAN

PyTorch/MLflow implementation of Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to perform single image super resolution (SISR) to downscale climate fields.
GNU General Public License v3.0
4 stars 1 forks source link
deep gan generative-adversarial-network learning pytorch

readme

DoWnGAN

DOwnscaling WassersteiN Generative Adversarial Network

This repo is under development as thesis work and is by not complete or tested.

This repo implements a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to perform single image super resolution (SISR) to downscale climate fields.

SISR is applied to ERA-Interim coarse input wind fields (but can be generalized to other fields) to acheive the same resolution as WRF U10 and V10 wind components.

Requirements

A working version of CUDA with Pytorch is required for this repo.

  1. Create a python virtual environment and activate it:

    python3 -m venv myvenv

    source myvenv/bin/activate

  2. Install requirements pip install -r requirements pip install "dask[distributed]"

  3. Install DoWnGAN (-e is essential if you want to customize the code) pip install -e /path/to/cloned/DoWnGAN/

Configuring the GAN

  1. Edit the paths in config/config.py

  2. To run the GAN, use the script in GAN/train.py python DoWnGAN/GAN/train.py

  3. Follow the instructions to log the file, and run a new mlflow server with python DoWnGAN/mlflow_tools/mlflow_server_cmd.py

  4. Go to localhost:5555 or specify port in mlflow_server_cmd.py file.

Check if CUDA is installed and PyTorch has access to cuda GPU

torch.cuda.is_available()

If the above returns True, then PyTorch is accessing the GPU.

References

Gulrajani, Ishaan, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. 2017. “Improved Training of Wasserstein GANs.” ArXiv:1704.00028 [Cs, Stat], December. http://arxiv.org/abs/1704.00028. Arjovsky, Martin, Soumith Chintala, and Léon Bottou. 2017. “Wasserstein GAN.” ArXiv:1701.07875 [Cs, Stat], December. http://arxiv.org/abs/1701.07875. Ledig, Christian, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, et al. 2017. “Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.” ArXiv:1609.04802 [Cs, Stat], May. http://arxiv.org/abs/1609.04802. Wang, Xintao, Ke Yu, Shixiang Wu, Jinjin Gu, Yihao Liu, Chao Dong, Chen Change Loy, Yu Qiao, and Xiaoou Tang. 2018. “ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks.” ArXiv:1809.00219 [Cs], September. http://arxiv.org/abs/1809.00219.