A Pytorch Implementation of PixelCNN++.

Main work taken from the official implementation

Pre-trained models are available here

I kept the code structure to facilitate comparison with the official code.

The code achieves 2.95 BPD on test set, compared to 2.92 BPD on the official tensorflow implementation.

python main.py

1. No data dependant weight initialization
2. No exponential moving average of past models for test set evalutation

For questions / comments / requests, feel free to send me an email.
Happy generative modelling :)

cd data
wget https://storage.googleapis.com/glow-demo/data/celeba-tfr.tar
tar -xvzf celeba-tfr.tar

Tensorflow needs to be installed to load the dataset, but the CPU version should be installed, ie pip install tensorflow to prevent TF from using GPU memory.