PU-Ray

IEEE | arXiv | Code | Data

Official implementation of "PU-Ray: Point Cloud Upsampling via Ray Marching on Implicit Surface".

Installation

Create virtual environment and install dependencies

conda create -n pu-ray python==3.8.17
conda activate pu-ray
pip install -r requirements.txt
pip install .

To check the installation:

pip freeze | grep pu-ray

Datasets

PU1K point clouds and PU-GAN mesh files are used for training and evaluation. Follow Grad-PU for data preparation.

Traning

To train with the prepared data, simply run the following command.

bash train.sh

Modify the arguments to experiment with different configurations. The command information can be found using the following command.

train_model -h
usage: train_model [-h] [--input-dir INPUT_DIR] [--query-dir QUERY_DIR]
                   [--log LOG] [--patch-k PATCH_K]
                   [--marching-steps MARCHING_STEPS] [--num-sample NUM_SAMPLE]
                   [--num-query NUM_QUERY] [--num-op NUM_OP] [--verbose]
                   [--seed SEED] [--num-epochs NUM_EPOCHS]

optional arguments:
  -h, --help            show this help message and exit
  --input-dir INPUT_DIR
                        Train data directory
  --query-dir QUERY_DIR
                        Query data directory
  --log LOG             log file name
  --patch-k PATCH_K     Patch size
  --marching-steps MARCHING_STEPS
                        Marching steps
  --num-sample NUM_SAMPLE
                        Number of training point clouds samples
  --num-query NUM_QUERY
                        Number of training query rays
  --num-op NUM_OP       Number of observation points
  --verbose             Print to stdout
  --seed SEED           Random seed
  --num-epochs NUM_EPOCHS
                        Number of epochs

Upsampling

Pre-trained models are provided. Run the following command to upsample the prepared point clouds.

bash upsample.sh

Modify the arguments to upsample with different configurations. The command information can be found using the following command.

pu_ray -h
usage: pu_ray [-h] [--input INPUT] [--output-dir OUTPUT_DIR] [--model MODEL] [--batch-size BATCH_SIZE] [--patch-k PATCH_K] [--query-k QUERY_K] [--r R] [--marching-steps MARCHING_STEPS] [--implicit-points]
              [--num-op NUM_OP] [--real-scanned] [--min-dist MIN_DIST] [--fps]

optional arguments:
  -h, --help            show this help message and exit
  --input INPUT         Input point cloud file name
  --output-dir OUTPUT_DIR
                        Name of the output file
  --model MODEL         Model name
  --batch-size BATCH_SIZE
                        Batch size
  --patch-k PATCH_K     Patch size
  --query-k QUERY_K     Neighbourhood size for novel query sampling
  --r R                 Upsampling rate
  --marching-steps MARCHING_STEPS
                        Marching steps
  --implicit-points     Output implicit points
  --num-op NUM_OP       Number of observation points
  --real-scanned        Real scanned data adaptation
  --min-dist MIN_DIST   Minimum distance to upsample (for real-scanned)
  --fps                 Real scanned data adaptation

Evaluation

Follow PU-GCN for evaluation of Chamfer Distance (CD), Hausdorff Distance (HD) and Point-to-Surface Distance (P2F).

References

@InProceedings{Qian_2021_CVPR,
    author    = {Qian, Guocheng and Abualshour, Abdulellah and Li, Guohao and Thabet, Ali and Ghanem, Bernard},
    title     = {PU-GCN: Point Cloud Upsampling Using Graph Convolutional Networks},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {11683-11692}
}

@inproceedings{li2019pugan,
     title={PU-GAN: a Point Cloud Upsampling Adversarial Network},
     author={Li, Ruihui and Li, Xianzhi and Fu, Chi-Wing and Cohen-Or, Daniel and Heng, Pheng-Ann},
     booktitle = {{IEEE} International Conference on Computer Vision ({ICCV})},
     year = {2019}
 }

@InProceedings{He_2023_CVPR,
    author    = {He, Yun and Tang, Danhang and Zhang, Yinda and Xue, Xiangyang and Fu, Yanwei},
    title     = {Grad-PU: Arbitrary-Scale Point Cloud Upsampling via Gradient Descent with Learned Distance Functions},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year      = {2023}
}

Citation

@misc{lim2023puray,
      title={PU-Ray: Point Cloud Upsampling via Ray Marching on Implicit Surface}, 
      author={Sangwon Lim and Karim El-Basyouny and Yee Hong Yang},
      year={2023},
      eprint={2310.08755},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}