[SIGGRAPH Asia 2022/TOG] The codes for 3QNet: 3D Point Cloud Geometry Quantization Compression Network published in ACM TRANSACTIONS ON GRAPHICS.

• TensorFlow 1.13.1
• Cuda 10.0
• Python 3.6.9
• numpy 1.14.5
• Sklearn
• Draco
• h5py
• bitarray
• Open3D

The adopted object models, indoor scenes, and outdoor scenes can be found in Visionair, Scannet, Kitti.

1. Compile

cd ./tf_ops
bash compile.sh

2. Train

Python3 3qnet.py

Note that the training h5 data should be put under ./data. The path for saved checkpoint files(savepath) should be edited according to your setting.

3. Test

Python3 test3q.py

Please put the ply files to be processed in indir, set the dracodir as where you install Draco, and edit modelpath as the path of checkpoint. The point clouds will be encoded into binary code files in bindir and decompressed into point clouds again into outdir. Note the compression level could be adjusted by changing level from 1 to 8.

Some decompressed samples under about 1bpp could be

If you find our work useful for your research, please cite:

@article{huang20223qnet,
  title={3QNet: 3D Point Cloud Geometry Quantization Compression Network},
  author={Huang, Tianxin and Zhang, Jiangning and Chen, Jun and Ding, Zhonggan and Tai, Ying and Zhang, Zhenyu and Wang, Chengjie and Liu, Yong},
  journal={ACM Transactions on Graphics (TOG)},
  volume={41},
  number={6},
  pages={1--13},
  year={2022},
  publisher={ACM New York, NY, USA}
}