This is an official implementation of Incremental Multiview Point Cloud Registration with Two-stage Candidate Retrieval that is accepted to Pattern Recognition.
Multiview point cloud registration serves as a cornerstone of various computer vision tasks. Previous approaches typically adhere to a global paradigm, where a pose graph is initially constructed followed by motion synchronization to determine the absolute pose. However, this separated approach may not fully leverage the characteristics of multiview registration and might struggle with low-overlap scenarios. In this paper, we propose an incremental multiview point cloud registration method that progressively aligns all scans to a growing meta-shape. This procedure increases the overlap between point cloud pairs, enhancing the overall multiview registration performance. To determine the incremental ordering, we employ a two-stage coarse-to-fine strategy for point cloud candidate retrieval. The first stage involves the coarse selection of scans based on neighbor fusion-enhanced global aggregation features, while the second stage further reranks candidates through geometric-based matching. Additionally, we apply a transformation averaging technique to mitigate accumulated errors during the registration process. Finally, we utilize a reservoir sampling-based technique to address density variance issues while reducing computational load. Comprehensive experimental results across various benchmarks validate the effectiveness and generalization of our approach.
First, create the conda environment.
conda create -n imvr python=3.8
conda activate imvr
pip install -r requirements.txt
Then, install the knn_search in ./model.
The code has been tested with Ubuntu 20.04, CUDA 11.7, Driver Version: 535.247.01, GeForce RTX 1080Ti, other platforms have not been tested but should work as well.
The data can be found from SGHR, the .pkl files can be found at SGHR/train/pkls, you need to remove the test_ prefix for three test .pkl files. Both our train and val folders correspond to the 3dmatch_train in SGHR, we provide a script in tools to help move the subfolders.
Please organize the data to ./data following the example data structure as:
data/
├── 3dmatch/
└── kitchen/
├── PointCloud/
├── cloud_bin_0.ply
├── gt.log
└── gt.info
├── yoho_desc/
└── 0.npy
└── Keypoints_PC_yoho/
└── cloud_bin_0Keypoints.txt
├── eth/
├── scannet/
├── train/
├── val/
├── 3dmatch.pkl
├── eth.pkl
├── scannet.pkl
├── train.pkl
└── val.pkl
If you want to use the FCGF descriptor, you need to extract the FCGF descriptor/keypoints and create two subfolders for FCGF under each scene.
data/
└── 3dmatch/
└── kitchen/
├── PointCloud/
├── yoho_desc/
└── 0.npy
├── fcgf_desc/
└── 0.npy
├── Keypoints_PC_fcgf/
└── cloud_bin_0Keypoints.txt
└── Keypoints_PC_yoho/
└── cloud_bin_0Keypoints.txt
Then generate the cache files for training by:
python tools/create_overlap_cache.py
The generated files will located in data/vlad_yoho_cache or data/vlad_fcgf_cache folder.
Train the overlap model.
python train_cache.py --config config/netvlad_yoho.yaml
We provide the pre-trained model checkpoints in release page, download and put the weight files to ./ckpt folder.
python test_imvr.py --config config/3dmatch_yoho_vlad.yaml
python test_imvr.py --config config/scannet_yoho_vlad.yaml
python test_imvr.py --config config/eth_yoho_vlad.yaml
python test_imvr.py --config config/3dmatch_fcgf_vlad.yaml
If you find this code useful for your work, please consider citing:
@article{li2025incremental,
title={Incremental multiview point cloud registration with two-stage candidate retrieval},
author={Li, Shiqi and Zhu, Jihua and Xie, Yifan and Zhu, Mingchen},
journal={Pattern Recognition},
pages={111705},
year={2025},
publisher={Elsevier}
}
We thank the authors of the LMVR, MultiReg, SGHR, FCGF for open sourcing their codes.