• Plese let me know if you find any issues or bugs: https://github.com/suga93/pixelcnn_keras/issues

• Version: 2.7.X or 3.4.X

• Version: 1.0.0rc1
• GPU: enabled

• Version: 1.2.2

• graphviz

• scipy==0.18.1
• numpy==1.12.0
• scikit-image==0.12.3
• h5py==2.6.0
• pydot-ng==1.0.0

This repository contains Dockerfile of execution environment. Docker image is registered using automated build.

• [tensorflow/tensorflow:1.0.0-rc1-devel-gpu-py3] (https://hub.docker.com/r/tensorflow/tensorflow/)

• suga93/pixelcnn_keras

• Linux OS
• GPU
• NVIDIA Driver
• Docker
• NVIDIA-Docker

1. Install Docker
2. Install NVIDIA-Docker
3. Get docker image: docker pull suga93/pixelcnn_keras
4. Run docker image: nvidia-docker run -it suga93/pixelcnn_keras

python3 train_keras_datasets.py -h

python3 train_keras_datasets.py -c configs/sample_train_mnist_small.cfg

python3 train_keras_datasets.py -c configs/sample_train_mnist_small.cfg --conditional True

python3 train_keras_datasets.py -c configs/sample_train_cifar10.cfg

• In this implementation, Keras fails to save model architecture (model.save(), model.to_json(), model.to_yaml()). Therefore, first you need to build same architecture as your trained model.

python3 predict_keras_datasets.py -h

python3 predict_keras_datasets.py -c configs/sample_predict_mnist.cfg --checkpoint_file /path/to/trained_model_mnist.hdf5

python3 predict_keras_datasets.py -c configs/sample_predict_cifar10.cfg --checkpoint_file /path/to/trained_model_cifar10.hdf5