smartdolphin / BEGAN-keras

A Keras implementation of the BEGAN Paper

BEGAN-keras

A Keras implementation of the Boundary Equilibrium GAN Paper
Warnning: This code is not working!

Extract faces data

Please extract image data in the faces folder.

$ unzip image_data.zip

Input data

└── began
    └── faces
        └── image_data
            └── xxx.png (name doesn't matter)

Usage

Install the necessary requirements. Import your data into the main.py file and make the appropriate parameter changes. The correctness of this implementation is still being verified.

Training locally

$ python main.py --batches_per_epoch=50 --batch_size=10 --image_dir=./out

Training in Cloud

This command is not yet supported

BUCKET_NAME=gs://{user}
TRAIN_DATA=gs://{user}/train_data
IMAGE_DIR=gs://{user}/out
JOB_NAME=began_train_$(date +%Y%m%d_%H%M%S); gcloud --verbosity=debug ml-engine jobs \
submit training $JOB_NAME \
--package-path=began --module-name=began.main \
--staging-bucket=$BUCKET_NAME --region=asia-east1 \
--config=began/cloudml-gpu.yaml \
-- --train_data=$TRAIN_DATA --image_dir=$IMAGE_DIR

Requirements

• Python 2.7 or 3
• Keras
• Numpy
• Matplotlib
• Tensorflow or Theano