Yue Pan 路 Tao Sun 路 Liyuan Zhu 路 Lucas Nunes 路 Iro Armeni 路 Jens Behley 路 Cyrill Stachniss
University of Bonn 路 Stanford University
- Release the inference code and RAP model v1.0.
- Release RAP model v1.1.
- Release the training code.
- Release the training data curation code and example training data.
- Add evaluation code on public datasets.
- Release RAP model v1.5 with other feature backbones, allowing metric scale difference, and handling 4D registration.
[Details (click to expand)]
Point cloud registration aligns multiple unposed point clouds into a common frame, and is a core step for 3D reconstruction and robot localization. In this work, we cast registration as conditional generation: a learned continuous, point-wise velocity field transports noisy points to a registered scene, from which the pose of each view is recovered. Unlike previous methods that conduct correspondence matching to estimate the transformation between a pair of point clouds and then optimize the pairwise transformations to realize multi-view registration, our model directly generates the registered point cloud. With a lightweight local feature extractor and test-time rigidity enforcement, our approach achieves state-of-the-art results on pairwise and multi-view registration benchmarks, particularly with low overlap, and generalizes across scales and sensor modalities. It further supports downstream tasks including relocalization, multi-robot SLAM, and multi-session map merging.Clone the repo:
git clone https://github.com/PRBonn/RAP.git
cd RAP
Setup conda environment:
conda create -n py310-rap python=3.10 -y
conda activate py310-rap
Install the dependency:
bash ./scripts/install.sh
Download model and example data:
bash ./scripts/download_weights_and_demo_data.sh
Try the demo by:
python app.py
Run batch inference after modifying the config files and the script test_script_example.sh:
bash ./scripts/test_script_example.sh
[Details (click to expand)]
If you use RAP for any academic work, please cite:
@article{pan2025arxiv,
title = {{Register Any Point: Scaling 3D Point Cloud Registration by Flow Matching}},
author = {Pan, Yue and Sun, Tao and Zhu, Liyuan and Nunes, Lucas and Armeni, Iro and Behley, Jens and Stachniss, Cyrill},
journal = arxiv,
volume = {arXiv:2512.01850},
year = {2025}
}
If you have any questions, please contact:
- Yue Pan {yue.pan@igg.uni-bonn.de}