Skip to content
ACM Transactions on Graphics (SIGGRAPH Asia 2018 issue), 2018
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
model Create Nov 30, 2018
LICENSE Create Nov 30, 2018



Deep Unsupervised Pixelization and Supplementary Material.
Chu Han^, Qiang Wen^, Shengfeng He*, Qianshu Zhu, Yinjie Tan, Guoqiang Han, and Tien-Tsin Wong (^joint first authors).
ACM Transactions on Graphics (SIGGRAPH Asia 2018 issue), 2018.

Our teaser


  • Python 3.5
  • PIL
  • Numpy
  • Pytorch 0.4.0
  • Ubuntu 16.04 LTS


Training Dataset

We collect 900 clip arts and 900 pixel arts for trianing our method. The folders named trainA and trainB contain the clip arts and pixel arts respectively here.

Testing Dataset

Create the folders testA and testB in the directory ./samples/. Note that testA and testB contain the clip arts to be pixelized and pixel arts to be depixelized respectively.


  • To train a model:
python3 ./ --dataroot ./samples --resize_or_crop crop --gpu_ids 0

or you can directly:

$ bash ./

You can check the losses of models in the file ./checkpoints_pixelization/loss_log.txt.
More training flags in the files ./options/ and ./options/


  • After training, all models have been saved in the directory ./checkpoints_pixelization/.
  • To test a model:
python3 ./ --dataroot ./samples --no_dropout --resize_or_crop crop --gpu_ids 0 --how_many 1 --which_epoch 200

or you can directly:

$ bash ./

More testing flags in the file ./options/
All testing results will be shown in the directory ./results_pixelization/.

You can’t perform that action at this time.