Manifold Fitting with CycleGAN

Pre-requisites

• Python 3.7+
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Examples can be found in the notebooks, which is self-contained and can be run independently.

Installation

• Clone this repo:

git clone https://https://github.com/zhigang-yao/MFCGAN
cd MFCGAN

• Install PyTorch and other dependencies (e.g., torchvision, visdom and dominate). You can do this by running the following:

pip install -r requirements.txt

MFCGAN train/test

• The data sets are included in the ./datasets folder.
• To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097.
• Train a model:

python train.py --dataroot ./datasets/circle --name circle

• Test the model, create the sub-networks and save the results:

python python create_test.py --dataroot ./datasets/circle --name circle

• The generators and data set will be save as .pth file in the folder ./results/circle/.

MFCGAN with images

• The data sets are included in the ./datasets folder.
• To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097.
• Train a model:

python train.py --dataroot ./datasets/gif/ --name gif --no_flip  --model vec2pic --dataset_mode vec2pic

• Test the model, create the sub-networks and save the results:

python create_test.py --dataroot ./datasets/gif/ --name gif_test --no_flip --model vec2pic --dataset_mode vec2pic --gpu_ids -1

• The generators and data set will be save as .pth file in the folder ./results/circle/.

Citation

If you use this code for your research, please cite our papers.

@article{yao2023manifold,
  title={Manifold Fitting with CycleGAN},
  author={Yao, Zhigang and Su, Jiaji and Yau, Shing-Tung},
  journal={Technical Report},
  year={2023}
}