DistGANs is a Python package to perform distributed training of conditional generative adversarial neural networks for multi-class labeled image data.
DistGANs partitionins the training data according to data labels, and enhances scalability by performing a parallel training where multiple generators are concurrently trained, each one of them focusing on a single data label.
This is a code implemented in collaboration with:
- Massimiliano Lupo Pasini at Oak Ridge National Laboratory (lupopasinim@ornl.gov)
- Nouamane Laanait at Oak Ridge National Laboratory (laanaitn@ornl.gov; nlaanait@gmail.com)
- Vittorio Gabbi at Politecnico di Milano (vittorio.gabbi@mail.polimi.it)
- Debangshu Mukherjee at Oak Ridge National Laboratory (mukherjeed@ornl.gov)
- Vitaliy Starchenko at Oak Ridge National Laboratory (starchenkov@ornl.gov)
- Junqi Yin at Oak Ridge National Laboratory (yinj@ornl.gov)
- Andrey Prokpenko at Oak Ridge National Laboratory (prokopenkoav@ornl.gov)
Python 3.5 or greater
PyTorch (any version works)
Optional, if NVIDIA gpu is present:
pip install pycuda
To keep similar code style, it should be formatted using black:
black -S -l 79 {source_file_or_directory}
conda create --name {env_name} python=3.7
conda install -n {env_name} matplotlib docopt ipython mpi4py
conda install -n {env_name} -c anaconda pyyaml
conda install -n {env_name} pytorch torchvision -c pytorch
conda install -n {env_name} tensorboardx -c conda-forge
All models should be located in GANs_dir. The class and the file that contains the class should have identic name. For example, class:
class CNN_model(GANs_abstract_object.GANs_model):
...
File:
CNN_model.py