Research map / Point clouds / Point cloud classification
PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Clouds
Yi Wei, Ziyi Wang, Yongming Rao, Jiwen Lu, Jie Zhou
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| combine point-based and voxel-based correlations (ablation) | PV-RAFT · FlyingThings3D | EPE (m): 0.0741 -> 0.0534 (point-only); 0.0712 -> 0.0534 (voxel-only) | Improved |
| add the refinement module (ablation) | PV-RAFT with point-voxel correlation · FlyingThings3D | EPE (m): 0.0534 -> 0.0461 | Improved |
| increase point KNN neighbors from 32 to 64 or 128 (ablation) | PV-RAFT point-correlation-only variant · FlyingThings3D | EPE (m): 0.0741 -> 0.2307; 0.0741 -> 0.6076 | Hurt |
| use a KNN pyramid with K=16,32,64 or K=32,64,128 (ablation) | PV-RAFT point-correlation variant · FlyingThings3D | EPE (m): 0.1616; 0.4841 | Hurt |
| vary voxel pyramid side length and depth (ablation) | PV-RAFT voxel-correlation variant · FlyingThings3D | EPE (m): 0.1408; 0.0902; 0.0712; 0.0672 | Mixed |
| use both point-voxel correlations instead of voxel correlation alone (ablation) | PV-RAFT without refinement · FlyingThings3D | EPE (m): 0.0712 -> 0.0534 | Improved |
| reduce truncation number to M=128 from M=512 (ablation) | PV-RAFT · FlyingThings3D | EPE (m): 0.0461 -> 0.0585 | Hurt |
| increase truncation number to M=1024 from M=512 (ablation) | PV-RAFT · FlyingThings3D | EPE (m): 0.0461 -> 0.0475 | No clear effect |
| replace point-voxel correlation with MLP+Maxpool correlation (ablation) | PV-RAFT correlation module · FlyingThings3D | EPE (m): 0.0461 -> 0.0704 | Hurt |
| replace point-voxel correlation with patch-to-patch correlation (ablation) | PV-RAFT correlation module · FlyingThings3D | EPE (m): 0.0461 -> 0.0614 | Hurt |
| use PV-RAFT instead of FLOT | scene flow estimation · FlyingThings3D | EPE (m): 0.052 -> 0.0461 | Improved |
| use PV-RAFT instead of FLOT | scene flow estimation · KITTI | Acc Strict: 0.755 -> 0.8226 | Improved |
Later papers that compared with this method
6 later accepted papers report PV-RAFT as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- GenFlow3D: Generative Scene Flow Estimation and Prediction on Point Cloud Sequences (ICCV 2025 · better on 2 of 2 shared results)
- milliFlow: Scene Flow Estimation on mmWave Radar Point Cloud for Human Motion Sensing (ECCV 2024 · better on 1 of 1 shared result)
- RMS-FlowNet++: Efficient and Robust Multi-scale Scene Flow Estimation for Large-Scale Point Clouds (IJCV 2024 · better on 1 of 1 shared result)
- PointConvFormer: Revenge of the Point-Based Convolution (CVPR 2023 · better on 2 of 3 shared results)
- NeuralPCI: Spatio-Temporal Neural Field for 3D Point Cloud Multi-Frame Non-Linear Interpolation (CVPR 2023 · better on 2 of 2 shared results)
- Multi-Scale Bidirectional Recurrent Network with Hybrid Correlation for Point Cloud Based Scene Flow Estimation (ICCV 2023 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D Point-Voxel Correlation Fields for Scene Flow Estimation (TPAMI 2023 · similarity 0.96)
- RAFT: Recurrent All-Pairs Field Transforms for Optical Flow (ECCV 2020 · similarity 0.88)
- RCP: Recurrent Closest Point for Point Cloud (CVPR 2022 · similarity 0.87)
- Efficient All-Pairs Correlation Volume Sampling for Optical Flow Estimation (CVPR 2026 · similarity 0.87)
- Self-Supervised 3D Scene Flow Estimation Guided by Superpoints (CVPR 2023 · similarity 0.87)
- SEA-Flow3D: Simplified, Efficient, and Accurate Scene Flow via Spatial Vector Sampling and Multi-scale Refinement (CVPR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
