The champion method HOUV about Regitstration on MVP Benchmark :* Multi-View Partial Point Clouds for Completion and Registration

[MVP Benchmark]

Overview

This repository mainly about champion method on the Partial-to-Partial Point Cloud Registration Completion. We propose a novel method called the Hybrid Optimization with Unconstrained Variables (HOUV) on 3D Registration. To the best of our knowledge, HOUV is the first work to directly optimize unconstrained variables in sophisticated registration, surpassing the advanced optimization and learning-based methods. This repository also introduces the MVP Benchmark for partial point cloud COMPLETION and REGISTRATION, and it also includes following recent methods:

• Completetion:   [1] PCN;   [2] ECG;   [3] VRCNet

• Registration:   [1] DCP;   [2] DeepGMR;   [3] IDAM

This repository is implemented in Python 3.7, PyTorch 1.5.0, CUDA 10.1 and gcc > 5.

Installation

Install Anaconda, and then use the following command:

git clone --depth=1 https://github.com/paul007pl/MVP_Benchmark.git
cd MVP_Benchmark; source setup.sh;

If your connection to conda and pip is unstable, it is recommended to manually follow the setup steps in setup.sh.

MVP Dataset

Download corresponding dataset:

• Completion :     Google Drive    or    百度网盘  (code: p364)
• Registration :     Google Drive    or    百度网盘  (code: p364)

Usage

For registration

• cd registration into the workspace.

• mkdir log and mkdir log/houv to build the savedir.

• touch log/houv/train.log to build the log file.

• To test MVP_ExtraTest_RG.h5 dataset: run bash run_test.sh.

  • This script needs four gpus.
  • It maybe take one or two hours.
  • If run error, please step by step commands in run_test.sh.

• Config for those methods can be found in cfgs/.

• The results are saved in ./log/houv/results.h5

• The submission in submission_last.zip More information please refer to registraion/read.me.

• Different partial point clouds for the same CAD Model:

• High-quality complete point clouds:

[Citation]

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

@article{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},
  journal={arXiv preprint arXiv:2104.10154},
  year={2021}
}

[License]

Our code is released under Apache-2.0 License.

[Acknowledgement]

We include the following PyTorch 3rd-party libraries:
[1] CD
[2] EMD
[3] MMDetection3D

We include the following algorithms:
[1] PCN
[2] ECG
[3] VRCNet
[4] DCP
[5] DeepGMR
[6] IDAM