A TensorFlow implementation of the PixelCNN.

There are three different datasets that the model is intended to be tested on: MNIST, Frey Face, and CIFAR-10 dataset.
To train and test the model, run the command: python main.py [--MNIST | --FREY | --CIFAR].

The model was trained for 25 epochs on a binarized version of the MNIST dataset.
The model was able to reach a negative log-likelihood score of 80.97 nats.

Incomplete Images:

Completed Images:

Incomplete Images:

Completed Images (Current):

Currently a work in progress...

The original PixelCNN paper was written by Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu.
The paper can be found here: Pixel Recurrent Neural Networks.