completion3D: Stanford 3D Object Point Cloud Completion Benchmark &

TopNet: Structural Point Cloud Decoder

This repository contains source code for all methods used for the Stanford 3D Object Point Cloud Completion Benchmark and presented in the paper TopNet: Structural Point Cloud Decoder, CVPR 2019.

Project Pages

The TopNet project page is available at https://completion3D.stanford.edu/topnet. The completion3D benchmark is available at http://completion3D.stanford.edu.

Overview

The completion3D benchmark is a platform for evaluating state-of-the-art 3D Object Point Cloud Completion methods. This repository contains source code for various methods evaluated on the benchmark. Both Tensorflow and Pytorch are supported. 3D Object Point Cloud Completion Results: A partial 3D point cloud is given as input and various methods used to generate a completed 3D point cloud

Benchmark submission instructions

To submit to the completion3d benchmark, set TRAIN=0 and BENCHMARK=1 in run.sh and run the script with parameters to evaluate. A submission.zip file will be generated by the script in the experiment output folder.

Citing this work

If you find this work useful in your research, please consider citing:

@inproceedings{topnet2019,
  title={TopNet: Structural Point Cloud Decoder},
  author={Tchapmi, Lyne P and Kosaraju, Vineet and Rezatofighi, S. Hamid and Reid, Ian and Savarese, Silvio},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}

@inProceedings{yuan2018pcn,
  title     = {PCN: Point Completion Network},
  author    = {Yuan, Wentao and Khot, Tejas and Held, David and Mertz, Christoph and Hebert, Martial},
  booktitle = {3D Vision (3DV), 2018 International Conference on},
  year      = {2018}
}

@article{DBLP:journals/corr/ChangFGHHLSSSSX15,
  author    = {Angel X. Chang and Thomas A. Funkhouser and Leonidas J. Guibas and Pat Hanrahan and Qi{-}Xing Huang and Zimo Li and Silvio Savarese and Manolis Savva and Shuran Song and Hao Su and Jianxiong Xiao and Li Yi and Fisher Yu},
  title     = {ShapeNet: An Information-Rich 3D Model Repository},
  journal   = {CoRR},
  volume    = {abs/1512.03012},
  year      = {2015},
  url       = {http://arxiv.org/abs/1512.03012},
  archivePrefix = {arXiv},
  eprint    = {1512.03012},
  timestamp = {Mon, 13 Aug 2018 16:47:39 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/ChangFGHHLSSSSX15},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

And please refer to the Shapenet Terms of Use

License

MIT License