A simple implementation of PixelCNN. It supports A/B style PixelCNN masking. It uses residual blocks. Image can be conditioned on a vector (eg class one hot vector).
Model in MNIST.ipynb achieved validation log likelihood 60.15 (nats)
Model in MNIST_ConditionedOnLabel.ipynb achieved valiation log likelihood 56.78 (nats)
References:
https://arxiv.org/abs/1601.06759
misc{oord2016pixel,
title={Pixel Recurrent Neural Networks},
author={Aaron van den Oord and Nal Kalchbrenner and Koray Kavukcuoglu},
year={2016},
eprint={1601.06759},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
https://arxiv.org/abs/1606.05328
misc{oord2016conditional,
title={Conditional Image Generation with PixelCNN Decoders},
author={Aaron van den Oord and Nal Kalchbrenner and Oriol Vinyals and Lasse Espeholt and Alex Graves and Koray Kavukcuoglu},
year={2016},
eprint={1606.05328},
archivePrefix={arXiv},
primaryClass={cs.CV}
}