MVP Benchmark: Partial-to-Partial Point Cloud Registration

We include the following methods for point cloud registration:

[1] DCP;   [2] DeepGMR;   [3] IDAM

MVP Registration Dataset

Download the MVP registration dataset Google Drive or 百度网盘  (code: p364) to the folder "data".

The data structure will be:

data
├── MVP_Train_RG.h5
|    ├── src (6400, 2048, 3)
|    ├── tgt (6400, 2048, 3)
|    ├── complete (6400, 2048, 3)
|    ├── cat_label (6400,)
|    ├── match_id (6400,)
|    └── match_level (6400,)
├── MVP_Test_RG.h5
|    ├── rotated_src (1200, 2048, 3)
|    ├── rotated_tgt (1200, 2048, 3)
|    ├── pose_src (1200, 4, 4)
|    ├── pose_tgt (1200, 4, 4)
|    ├── rot_level (1200,)
|    ├── transforms (1200, 4, 4)
|    ├── src (1200, 2048, 3)
|    ├── tgt (1200, 2048, 3)
|    ├── complete (1200, 2048, 3)
|    ├── cat_label (1200,)
|    ├── match_id (1200,)
|    └── match_level (1200,)
└── MVP_ExtraTest_RG.h5
     ├── rotated_src (2000, 2048, 3)
     ├── rotated_tgt (2000, 2048, 3)
     └── cat_label (2000,)

Attention: MVP_ExtraTest_Shuffled_RG.h5 is only available during the competition period. MVP_Test_RG.h5 is the standard test set if you use MVP dataset in your papers.

We create the registration dataset by ensuring surfficent overlaps between the source point cloud and the target. Partial point cloud pairs with "match_level = 1" mostly have more correspondences than those with "match_level = 0".

Most relative rotations are within [0, 45\textdegree], and the rest have unrestricted rotations \in [0, 360\textdegree]. The ratio is roughly 4 : 1.

Note that the source and the target are two different incomplete point clouds scanned from the same CAD model.

Usage

• To train a model: run python train.py -c ./cfgs/*.yaml, e.g. python train.py -c ./cfgs/pcn.yaml
• To test a model: run python test.py -c ./cfgs/*.yaml, e.g. python test.py -c ./cfgs/pcn.yaml
• Config for each algorithm can be found in cfgs/.
• run_train.sh and run_test.sh are provided for SLURM users.

Citation

If you find our code useful, please cite our paper:

@inproceedings{pan2021variational,
  title={Variational Relational Point Completion Network},
  author={Pan, Liang and Chen, Xinyi and Cai, Zhongang and Zhang, Junzhe and Zhao, Haiyu and Yi, Shuai and Liu, Ziwei},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={8524--8533},
  year={2021}
}
@article{pan2021robust,
  title={Robust Partial-to-Partial Point Cloud Registration in a Full Range},
  author={Pan, Liang and Cai, Zhongang and Liu, Ziwei},
  journal={arXiv preprint arXiv:2111.15606},
  year={2021}
}
@article{pan2021mvp, 
  title={Multi-View Partial (MVP) Point Cloud Challenge 2021 on Completion and Registration: Methods and Results}, 
  author={Pan, Liang and Wu, Tong and Cai, Zhongang and Liu, Ziwei and Yu, Xumin and Rao, Yongming and Lu, Jiwen and Zhou, Jie and Xu, Mingye and Luo, Xiaoyuan and Fu, Kexue, and Gao, Peng, and Wang, Manning, and Wang, Yali, and Qiao, Yu, and Zhou, Junsheng, and Wen, Xin, and Xiang, Peng, and Liu, Yu-Shen, and Han, Zhizhong, and Yan, Yuanjie, and An, Junyi, and Zhu, Lifa, and Lin, Changwei, and Liu, Dongrui, and Li, Xin, and G ́omez-Fern ́andez, Francisco, and Wang, Qinlong, and Yang, Yang}, 
  journal={arXiv preprint arXiv:2112.12053},
  year={2021}
}

License

Our code is released under Apache-2.0 License.

Acknowledgement

We include the following algorithms:
[1] DCP
[2] DeepGMR
[3] IDAM