ToniCreswell / attribute-cVAEGAN

Conditional Autoencoders with Adversarial Information Factorization

Conditional Autoencoders with Adversarial Information Factorization

(New version of the code using ResNets will be provided soon to reproduce results in the latest verision of our paper: https://arxiv.org/pdf/1711.05175.pdf)

Updates to our paper and code (coming soon) include:

1. Classification results on CelebA facial attributes
2. Use of ResNets to improve image quality

To use code:

1. Download the celebA dataset from http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
2. Install dependancies listed in requirements.txt
3. You will also need pyTorch which may be downloaded from http://pytorch.org
4. Run the jupyter notebooks to get the data tensors xTrain.npy and yAllTrain.npy and move them in to folder celebA/InData/
5. The code may be run from cmd line with various options detailed in the code

Example results: