Fake-Face-Generation

This is the pytorch implementation of 3 different GAN models using same convolutional architecture.

• DCGAN (Deep convolutional GAN)
• WGAN-CP (Wasserstein GAN using weight clipping)
• WGAN-GP (Wasserstein GAN using gradient penalty)

Dependencies

• Python 3.x
• PyTorch
• torchvision
• numpy
• matplotlib
• tqdm

Usage

1. Clone the repository:

git clone https://github.com/sherlockholmes1603/Fake-Face-Generation.git

2. Install dependencies:

pip install -r requirements.txt

3. Train the GAN models:

python3 main.py --model DCGAN/WGAN-CP/WGAN-GP --is-train --dataroot "/kaggle/input/celeba-dataset/img_align_celeba/" --epochs 10 --cuda --load_G models/wgan_cp/generator_epoch5.pth --load_D models/wgan_cp/critic_epoch5.pth 

4. Generate fake faces:

python3 main.py --model DCGAN/WGAN-CP/WGAN-GP  --load_G models/wgan_cp/generator_epoch5.pth --load_D models/wgan_cp/critic_epoch5.pth 

Note:

Replace DCGAN, WGAN-CP, or WGAN-GP with the desired model. --dataroot should point to the CelebA dataset. --epochs is the number of training epochs. --load_G and --load_D are the paths to the pre-trained generator and discriminator models, respectively. Once the training is complete, the generated fake faces will be saved in the results directory.

Fake images generated by the models are

1. DCGAN

2. WGAN with weight clipping

3. WGAN with gradient penalty

• The implementation is inspired by the original papers on DCGAN, WGAN-CP, and WGAN-GP.
• Dataset used for training: [link to dataset]