/Rethinking-CGANs

Primary LanguagePythonApache License 2.0Apache-2.0

Implementation of "Rethinking Conditional GAN training: an approach using geometrically structured latent manifolds." [PDF]

alt text

Requirements

Tensorflow = 1.9 Python = 3.x Scipy = 1.4

Data preparation

The training files should be placed at data_dir/images/training/

The validation files should be placed at data_dir/images/validation/

Each training and validation file should be a concatination of the input and the corresponding output. An example is shown below:

alt text

Train

python p2p_geo.py --mode=train --data_dir=data directory --learning_rate=1-e4 --logs_dir=logs directory --batch_size=25

Visualize

python p2pGeo.py --mode=visualize