Redirecting to original paper in 30 seconds...
Click below to go immediately or wait for automatic redirect
📄 Abstract
Abstract: We introduce Rectified Point Flow, a unified parameterization that formulates
pairwise point cloud registration and multi-part shape assembly as a single
conditional generative problem. Given unposed point clouds, our method learns a
continuous point-wise velocity field that transports noisy points toward their
target positions, from which part poses are recovered. In contrast to prior
work that regresses part-wise poses with ad-hoc symmetry handling, our method
intrinsically learns assembly symmetries without symmetry labels. Together with
a self-supervised encoder focused on overlapping points, our method achieves a
new state-of-the-art performance on six benchmarks spanning pairwise
registration and shape assembly. Notably, our unified formulation enables
effective joint training on diverse datasets, facilitating the learning of
shared geometric priors and consequently boosting accuracy. Project page:
https://rectified-pointflow.github.io/.
Authors (5)
Tao Sun
Liyuan Zhu
Shengyu Huang
Shuran Song
Iro Armeni
Key Contributions
Rectified Point Flow introduces a unified formulation for point cloud registration and shape assembly as a single conditional generative problem. It learns a continuous point-wise velocity field to transport noisy points to target positions, intrinsically learning assembly symmetries without explicit labels. This approach achieves state-of-the-art performance on multiple benchmarks by enabling effective joint training on diverse datasets.
Business Value
Enables more accurate and efficient 3D reconstruction and object manipulation for applications in robotics, autonomous driving, and AR/VR, leading to improved automation and user experiences.