README.md

---
license: apache-2.0
pipeline_tag: robotics
tags:
- point-cloud
- 3d-vision
- pose-estimation
- registration
- flow-model
- computer-vision
---

# [NeurIPS'25] Rectified Point Flow: Generic Point Cloud Pose Estimation

[![Project Page](https://img.shields.io/badge/Project_Page-RPF-blue)](https://rectified-pointflow.github.io/) 
[![arXiv](https://img.shields.io/badge/arXiv-2506.05282-blue?logo=arxiv&color=%23B31B1B)](https://arxiv.org/abs/2506.05282) 
[![GitHub](https://img.shields.io/badge/GitHub-Repository-black?logo=github)](https://github.com/GradientSpaces/Rectified-Point-Flow)


**Rectified Point Flow (RPF)** is a unified model that formulates pairwise point cloud registration and multi-part shape assembly as a single conditional generative problem. Given unposed point clouds, the method learns a continuous point-wise velocity field that transports noisy points toward their target positions, from which part poses are recovered.


## Installation
```bash
git clone https://github.com/GradientSpaces/Rectified-Point-Flow.git
cd Rectified-Point-Flow
conda create -n py310-rpf python=3.10 -y
conda activate py310-rpf
poetry install  # or `uv sync`, `bash install.sh`
```

## Quick Start
```bash
# Assembly Generation:
python sample.py data_root=./demo/data

# Overlap Prediction:
python predict_overlap.py data_root=./demo/data
```
More details can be found in our [GitHub Repo](https://github.com/GradientSpaces/Rectified-Point-Flow).

## Checkpoints
- `RPF_base_full_*.ckpt`: Complete model checkpoint for assembly generation
- `RPF_base_pretrain_*.ckpt`: Encoder-only checkpoint for overlap prediction

## Training Data
| Dataset | Task | Part segmentation source | Parts per sample |
|---|---|---|---|
| [**IKEA-Manual**](https://yunongliu1.github.io/ikea-video-manual/) | Shape Assembly | Defined by IKEA manuals | [2, 19] |
| [**PartNet**](https://partnet.cs.stanford.edu/) | Shape Assembly | Human-annotated parts | [2, 64] | 
| [**BreakingBad-Everyday**](https://breaking-bad-dataset.github.io/) | Shape Assembly | Simulated fractures via [fracture-modes](https://github.com/sgsellan/fracture-modes#dataset) | [2, 49] |
| [**Two-by-Two**](https://tea-lab.github.io/TwoByTwo/) | Shape Assembly | Annotated by human | 2 |
| [**ModelNet-40**](https://github.com/GradientSpaces/Predator) | Pairwise Registration | Following [Predator](https://github.com/prs-eth/OverlapPredator) split | 2 |
| [**TUD-L**](https://bop.felk.cvut.cz/datasets/) | Pairwise Registration | Real scans with partial observations | 2 |
| [**Objverse**](https://objaverse.allenai.org/) | Overlap Prediction | Segmented by [SAMPart3D](https://github.com/GradientSpaces/SAMPart3D) | [3, 12] |

## Citation
```bibtex
@inproceedings{sun2025_rpf,
    author = {Sun, Tao and Zhu, Liyuan and Huang, Shengyu and Song, Shuran and Armeni, Iro},
    title = {Rectified Point Flow: Generic Point Cloud Pose Estimation},
    booktitle = {NeurIPS},
    year = {2025},
}
```