Source code :
src/
datasets.py - Different datasets classes that share common interface, it loads data and provides generators
to iterate over minibatches
model_***.py - Classes that implement AAE architectures. Dense neural networks, convolution neural networks,
networks that use Subpix layers.
aae_**_solver.py - Classes that implement different training procedures. Pixel Matching AAE, Feature matching
AAE using GAN network
utils.py - Functions providing higher level abstraction over pure tensorflow ops.
train_aae.py - Training procedure with different scenarios
interface.py - Little GUI program, used for sampling from models and traversing latent space
draw_samples.py - Visualize samples from input encoded into latent representation
draw_images_**.py - Scripts to generate images from models produced during training.
create_*.py - Scripts used to create datasets
Models and Images: https://drive.google.com/drive/folders/0B2ZqB_V870aATHF5UmVrR1VCZGc?usp=sharing
Note:
Mnist dataset will be downloaded automatically
Celeb dataset has to be added manually to the CELEB folder