[TVCG 2021] Consistent Two-Flow Network for Tele-Registration of Point Clouds
We present a learning-based technique that allows registration between point clouds presented in arbitrary poses, and having little or even no overlap, a setting that has been referred to as tele-registration.
Our technique is based on a novel neural network design that learns a prior of a class of shapes and can complete a partial shape. The key idea is combining the registration and completion tasks in a way that reinforces each other. In particular, we simultaneously train the registration network and completion network using two coupled flows, one that register-and-complete, and one that complete-and-register, and encourage the two flows to produce a consistent result.
For more details, please refer to our project page.
The training data are generated by virtual scanning and sphere cropping.
Specifically, given a mesh model, the point cloud data is first obtained by virtual scanning, then, the partial point cloud pair is generated by random sphere cropping (see crop_pc.py in data folder). Finally, all the data are organized in HDF5 format (see create_h5.py in data folder).
In our implementation, we select 8 categories in the ShapeNet v2 dataset to create our training data, the train/validate/test list is provided in data folder.
The codes are tested under PyTorch 1.7.0 GPU version and Python 3.7 on Linux Mint 20.
First, please install the EMD module.
Run train.py to train CTF-Net, the trained models will be saved in CTF-Net/output/ folder.
Run test.py to test the trained models.
Our code is released under MIT License. See LICENSE file for details.
Please cite the paper in your publications if it helps your research:
@article{CTFNet21,
title={Consistent Two-Flow Network for Tele-Registration of Point Clouds},
author={Zihao Yan and Zimu Yi and Ruizhen Hu and Niloy J. Mitra and Daniel Cohen-Or and Hui Huang},
journal={IEEE Transactions on Visualization and Computer Graphics},
volume={},
pages={},
year={2021},
doi={10.1109/TVCG.2021.3086113}
}