Encoding people's faces with autoencoders. All experiments were conducted on Labeled Faces in the Wild

Files description

• gifs/* - animated reconstruction histories for validation images
• config.py -- various parameters (eg., image size, model architectures)
• get_dataset.py -- helper module for downloading and preprocessing data
• data_utils.py -- further preprocessing, sampling and plotting functions
• modules.py -- helper modules for restoring size, padding and unified output for autoencoders
• losses.py -- redefined MSE for unified training and VAE loss
• train_util.py -- model training while saving losses and reconstructions history; animation generation, t-SNE projections and search for similar images are also implemented here
• base_ae.py -- abstract base class from which other autoencoders inherit to
• vanilla_ae.py -- simple autoencoder with few dense layers
• vae.py - convolutional VAE and CVAE (two in one)
• demo.ipynb -- AE, VAE and CVAE demo with applications