# High-Quality Multi-View Partial Point Cloud for Completion

Liang Pan1      Xinyi Chen1 2      Zhongang Cai2 3      Junzhe Zhang1 2
Haiyu Zhao2 3      Shuai Yi2 3      Ziwei Liu1

Multi-View Partial (MVP) Point Cloud Dataset. (a) shows an example for our 26 uniformly distributed camera poses on a unit sphere. (b) presents the 26 partial point clouds for the airplane from our uniformly distributed virtual cameras. (c) compares the rendered incomplete point clouds with different camera resolutions. (d) shows that Poisson disk sampling generates complete point clouds with a higher quality than uniform sampling.

## News

2021-07-12 The submission on Codalab starts!

2021-07-10 The benchmark, "Single-View Point Cloud Completion" has been released.

2021 The MVP challenges will be hosted in the ICCV2021 Workshop: Sensing, Understanding and Synthesizing Humans

.

## Details

Single-View Point Cloud Completion Benchmark evaluates the performance of generating complete 3D point clouds based on single-view partial point clouds. This is a large subset of MVP, containing massive high-quality incomplete and complete point clouds generated from 3D CAD models. It contains 163,800 number of partial point clouds and 6,300 number of complete point clouds from 16 number of object categories:

• The Training set has 62,400 point cloud pairs, and their corresponding category labels.

• The Testing set has 41,600 point cloud pairs, and their corresponding category labels.

• The ExtraTesting set has 59,800 partial point clouds, and their corresponding category labels. We keep the groundtruth complete point clouds, and you can evaluate your method by submitting your results on the Codalab website.

The general pipeline is training your method on the Training set while evaluating on the Testing set. Finally, you can generate your results for the ExtraTest set by your best model. Note that the Training set and Testing set are the same with the data (2048 points) prepared in the paper VRCNet. Using the Testing set for training is strictly forbidden.

## Citation

@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}
}