GANTransferLimitedData

This is a pytorch implementation of the paper On Leveraging Pretrained GANs for Limited-Data Generation.

Please consider citing our paper if you refer to this code in your research.

@inproceedings{zhao2020leveraging,
  title={On Leveraging Pretrained GANs for Limited-Data Generation},
  author={Zhao, Miaoyun and Cong, Yulai and Carin, Lawrence},
  booktitle={ICML},
  year={2020},
}

Requirement

python=3.7.3
pytorch=1.2.0

Notes

CELEBA_[f]GmDn.py is the implementation of the model in Figure1(f).

Flower_[h]our.py is the implementation of the model in Figure1(h). This code is for "Section 4.1 Comparisons with Existing Methods".

Flower25_our.pyis the code for the experiments on Flowers-25.

Usage

First, download the pretrained GP-GAN model by running download_pretrainedGAN.py. Note please change the path therein.

Second, download the training data to the folder ./data/. For example, download the Flowers dataset from: https://www.robots.ox.ac.uk/~vgg/data/flowers/102/ to the folder ./data/102flowers/. For Flowers-25, we choose the first 25 images from the passion category, following Image Generation from Small Datasets via Batch Statistics Adaptation.

Dataset preparation

data
├──102flowers
           ├──all8189images
                      ├──image_folder
           ├──passion25 
                      ├──image_folder
├── CelebA
...

Finally, run Flower_[h]our.py or Flower25_our.py.

Acknowledgement

Our code is based on GAN_stability: https://github.com/LMescheder/GAN_stability from the paper Which Training Methods for GANs do actually Converge?.