Point Cloud Compression with Sibling Context and Surface Priors

Point Cloud Compression with Sibling Context and Surface Priors
Zhili Chen, Zian Qian, Sukai Wang, Qifeng Chen
European Conference on Computer Vision (ECCV) 2022

[Abs][Paper]

This repository is an official implementation of "Point Cloud Compression with Sibling Context and Surface Priors" (PCC-S). This is the initial version of PCC-S and this repo will be keep updated.

Preparation

Environment

conda create --name PCC-S python=3.8
conda activate PCC-S
sudo apt install build-essential python3-dev libopenblas-dev
conda install pytorch cudatoolkit -c pytorch
export CUDA_HOME=/usr/local/cuda-11.4; # or select the correct cuda version on your system.
git clone https://github.com/NVIDIA/MinkowskiEngine.git
cd MinkowskiEngine
python setup.py install --blas_include_dirs=${CONDA_PREFIX}/include --blas=openblas
# To the root of the project folder
cd ..
bash requirements.sh
git submodule add https://github.com/ThibaultGROUEIX/ChamferDistancePytorch

Data

1. Download the LiDAR point cloud data from Kitti Odometry Dataset.

2. Set the absolute paths for variables ROOT_dir, KITTI_BIN_dir, and folder name for variable PROCESSED_DATA_dirfor saving the preprocessed data in config_ent.yml.

   • ROOT_dir: Absolute path to this repo.

   • KITTI_BIN_dir: Absolute path to KITTI point cloud sequence.

   • PROCESSED_DATA_dir: Folder name for saving the preprocessed data.

3. Preprocess data with the following command.

python util/preprocess_data.py --n_workers 16

Compression

• Train the entropy model. Set the relative path for variable of CKPT_DIR to save checkpoint in config_ent.yml.

CUDA_VISIBLE_DEVICES='0,1' python PCCS.py

• Evaluate the compression performance. Set the folder name for variable CKPT_DIR to the trained checkpoint of the entropy model in config_ent.yml or set as pretrained to use the provided pretrained model.

CUDA_VISIBLE_DEVICES='0' python PCCS.py --is_validation

Refinement

1. Set folder name for variable pre_cache_feats_dir in config_ent.yml for saving the cached predicted information from the entropy model. Then run the following command.

   CUDA_VISIBLE_DEVICES='0' python PCCS.py --to_cache_par_feats

2. Train the refinement model. Set the folder name for variable CKPT_DIR to save checkpoint in config_refine.yml.

   CUDA_VISIBLE_DEVICES='0,1' python refine.py

3. Evaluate the reconstruction performance. Set the folder name for variable CKPT_DIR to the trained checkpoint of the entropy model in config_refine.yml or set as pretrained to use the provided pretrained model.

   CUDA_VISIBLE_DEVICES='0' python refine.py --is_validation

Encode and Decode

• Encode and decode. Set compress_save_dir in config_ent.yml for saving the encoded files and reconstructed point cloud in the form of ply .

CUDA_VISIBLE_DEVICES='0' python Encode_Decode.py

Citation

If you find this project useful for your research, please consider citing:

@article{chen2022point,
  title={Point Cloud Compression with Sibling Context and Surface Priors},
  author={Chen, Zhili and Qian, Zian and Wang, Sukai and Chen, Qifeng},
  journal={arXiv preprint arXiv:2205.00760},
  year={2022}
}

Contact

Feel free to contact me if there is any question. (Zhili Chen, [email protected] )