VCR-Net: Visual Corresponding for Registraion

Quick Visualization

This is the our IROS 2020 work. VCR-Net is a deep-learning approach designed for performing rigid partial-partial point cloud registration. Our paper can be found on Arxiv.

@INPROCEEDINGS{9341249,  author={Wei, Huanshu and Qiao, Zhijian and Liu, Zhe and Suo, Chuanzhe and Yin, Peng and Shen, Yueling and Li, Haoang and Wang, Hesheng},  booktitle={2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},   title={End-to-End 3D Point Cloud Learning for Registration Task Using Virtual Correspondences},   year={2020},  volume={},  number={},  pages={2678-2683},  doi={10.1109/IROS45743.2020.9341249}}

Prerequisites

sympy 
h5py 
tqdm 
tensorboardX  
torchvision==0.7.0 
pytorch==1.6.0

whole to whole

Train

#pre-train
python main.py --dataset=modelnet40 --test_batch_size=16 --batch_size=16 --model=lpd
#train
python main.py --dataset=modelnet40 --test_batch_size=16 --batch_size=4 -model_path=./pretrained/lpd-pretrained.t7 

Test

python main.py --dataset=modelnet40 --test_batch_size=16 --batch_size=4 --model_path=./pretrained/vcrnet-whole.t7 --eval

part to part

Train

python main.py --dataset=modelnet40 --test_batch_size=24 --batch_size=4  --partial  --overlap=0.575 ---model_path=./pretrained/vcrnet-whole.t7

Test

python main.py --dataset=modelnet40 --test_batch_size=24 --batch_size=4 --partial  --overlap=0.575 --model_path=./pretrained/vcrnet-part.t7 --iter=3 --eval

Thanks

DCP

Zhuowen Shen