This repo includes the implementation on pytorch of a fully deep convolutional generative adversial network,
that tries to create realistic faces of the CelebA datastet. The architecture and model
hyperparameters are based on the paper
"Unsupervised representation learning with DCGANs" (https://arxiv.org/abs/1511.06434)
It includes two different models based on the same architecture:
- DCGAN
- WGAN
The second based on the paper "Wasserstein GAN" (https://arxiv.org/abs/1701.07875)
The realism of the images generated by the models is limited by my processing resources
Sample image of the DCGAN trained on 23 epochs Sample image of the WGAN trained on 215000 epochs Here I made vector arithmetic with the latent vectors of the generated images to make a neutral man smile
The dataset is available at http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
Pol Monroig