Progressive Autoencoder

Progressive training of Autoencoder for the generation of 3D MRI brains.

Installation

For the installation of the environment :

  $ pip install -r requirements.txt 

Training

To progressivly train the AE using pretrained generators :

  $ python3 main.py train --generator_folder mygenerators/ --save_folder mymodels/ --stop_res 4 

Required arguments are:

• generator_folder : folder to the .h5 pre-trained models to be used as the image generator & AE decoder
• save_folder : folder where trained AE should be stored
• stop_res : final image resolution to achieve during training (e.g: 4, 8, 16, 32, 64, 128, 256)

Additional arguments :

• start_res : initial image resolution to start training with (default: 2)
• latent : latent code size (default: 1024)

Testing

Not Implemented Yet