ADT-GAN program, the source code for our submitted paper "ADT-GAN: Adversarial and Discriminative Learning for Transferring Generative Adversarial".

Requirements

tensorflow 1.X
keras
torch
numpy
matplotlib
opencv-python
tqdm

Install the needed packages for running ADT-GAN program

When using CPU:
    $ pip install -r requirements.txt

When using GPU:
    $ pip install -r requirements_gpu.txt

Function of each directory

1. 0.Pre-DCGAN: You need to pre-train on dataset MNIST-not9 with Pre-DCGAN, and Copy models G1 and D1 to models. We have saved the pre-trained models G1 and D1 that you can use directly.

2. 1.DCGAN: Training DCGAN on MNIST by 1.DCGAN/DCGAN_MNIST. Training DCGAN on CelebA by 1.DCGAN/DCGAN_CelebA.

3. 2.Initialized-DCGAN: Training Initialized-DCGAN on MNIST by 2.Initialized-DCGAN/Initialized-DCGAN_MNIST. Training Initialized-DCGAN on CelebA by 2.Initialized-DCGAN/Initialized-DCGAN_CelebA.

4. 3.ADT-GAN: Training ADT-GAN on MNIST by 3.ADT-GAN/ADT-GAN_MNIST. Training ADT-GAN on CelebA by 3.ADT-GAN/ADT-GAN_CelebA.

5. FID: The FID is the performance measure used to evaluate the experiments in the paper.

Please see the README.md under each of the above directory for guidance on running the program.