One Framework to Register Them All: PointNet Encoding for Point Cloud Alignment

[Paper]

Source Code Author: Vinit Sarode, Xueqian Li and Animesh Dhagat

Requirements:

1. Cuda 10
2. tensorflow==1.14
3. transforms3d==0.3.1
4. h5py==2.9.0
5. pytorch==1.3.0

Dataset:

Path for dataset: Link

1. Download 'train_data' folder from above link.
2. Download 'car_data' folder from above link.

Point Cloud Registration Network:

Train Iterative-PCRNet:

1. cd pcrnet
2. chmod +x train_itrPCRNet.sh
3. ./train_itrPCRNet.sh

Note: PCRNet's pytorch implementation is also available at [Link].

PointNetLK:

Train PointNetLK:

1. cd pnlk
2. cd experiments
3. python train_pointlk.py

Citation

@misc{sarode2019framework,
    title={One Framework to Register Them All: PointNet Encoding for Point Cloud Alignment},
    author={Vinit Sarode and Xueqian Li and Hunter Goforth and Yasuhiro Aoki and Animesh Dhagat and Rangaprasad Arun Srivatsan and Simon Lucey and Howie Choset},
    year={2019},
    eprint={1912.05766},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}