Rectified Point Flow: Generic Point Cloud Pose Estimation
This work addresses the challenge of accurately estimating poses for point clouds in computer vision and robotics, with incremental improvements in handling symmetries and joint training.
The paper tackles the problem of point cloud pose estimation for registration and shape assembly by introducing Rectified Point Flow, a unified conditional generative approach that learns a continuous velocity field to transport points to target positions, achieving state-of-the-art performance on six benchmarks.
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/.