Code for Differential-Critic GAN: Generating What You Want by a Cue of Preferences (TNNLS 2022)

• python3
• pytorch
• NumPy, SciPy, Matplotlib

1. Pretrain mnist classifier:

• Under ./cls_mnist dir: run python cls_mnist.py --save-model
  • Expected Output
    • mnist_cnn.pt

2. Pretrain WGAN:

• Run python gan_mnist.py
  • Expected Output
    • results/pretrain

3. Train DiCGAN:

• Run python dicgan_mnist.py
  • Expected Output
    • results/$RUN_NAME/0/positive.txt records the percentage of desired samples at each epoch
    • results/$RUN_NAME/0/samples contain sampled outputs from generator
    • Default $RUN_NAME is dicgan_small_digits

@article{yao2022differential,
title={Differential-Critic GAN: Generating What You Want by a Cue of Preferences},
author={Yao, Yinghua and Pan, Yuangang and Tsang, Ivor W and Yao, Xin},
journal={IEEE Transactions on Neural Networks and Learning Systems},
year={2022},
publisher={IEEE}
}