A Keras implementation of the Boundary Equilibrium GAN Paper
Warnning: This code is not working!
Please extract image data in the faces folder.
$ unzip image_data.zip
Input data
└── began
└── faces
└── image_data
└── xxx.png (name doesn't matter)
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
- Python 2.7 or 3
- Keras
- Numpy
- Matplotlib
- Tensorflow or Theano