Skip to content

Latest commit

 

History

History
29 lines (21 loc) · 1.28 KB

readme.md

File metadata and controls

29 lines (21 loc) · 1.28 KB

PixelCNN++

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.

Running the code

python main.py

Differences with official implementation

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

Contact

This repository is no longer maintained. Feel free to file an issue if need be, however response may be slow.