3D Point-Voxel Correlation Fields for Scene Flow Estimation 

TPAMI 2023

Ziyi Wang* , Yi Wei*, Yongming Rao, Jie Zhou, Jiwen Lu

Department of Automation, Tsinghua University, China

[Paper (IEEE)]      [Code (GitHub)]


we propose Point-Voxel Correlation Fields to explore relations between two consecutive point clouds and estimate scene flow that represents 3D motions.

To exploit point-based correlations, we adopt the K-Nearest Neighbors search that preserves fine-grained information in the local region, which guarantees the scene flow estimation precision. 

By voxelizing point clouds in a multi-scale manner, we construct pyramid correlation voxels to model long-range correspondences, which are utilized to handle fast-moving objects. 

Point-Voxel Recurrent All-Pairs Field Transforms (PV-RAFT)

Deformable PV-RAFT (DPV-RAFT)

In DPV-RAFT, we propose the Spatial Deformation that deforms the voxelized neighborhood in the voxel correlation branch.

In DPV-RAFT, we propose the Temporal Deformation that  controls the iterative update process.

Experiment Results

BibTeX

@article{wang2023dpvraft,

  title={3D Point-Voxel Correlation Fields for Scene Flow Estimation},

  author={Wang, Ziyi and Wei, Yi and Rao, Yongming and Zhou, Jie and Lu, Jiwen},

  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},

  year={2023},

  publisher={IEEE}

}