Research map / Point clouds / Point cloud classification
FlowNet3D: Learning Scene Flow in 3D Point Clouds
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Estimate scene flow end-to-end from point clouds using FlowNet3D | FlowNet3D · FlyingThings3D test | EPE: 0.3401 -> 0.1694 | Helped |
| Estimate scene flow end-to-end from point clouds using FlowNet3D | FlowNet3D · KITTI scene flow (without ground points) | 3D EPE: 0.215 -> 0.122 | Helped |
| Use max pooling instead of average pooling (ablation) | FlowNet3D flow embedding · FlyingThings3D test | EPE: 0.3163 -> 0.2463 | Helped |
| Use learned feature relations instead of dot-product feature distance (ablation) | FlowNet3D flow embedding · FlyingThings3D test | EPE: 0.2463 -> 0.2298 | Helped |
| Use learned feature relations instead of cosine feature distance (ablation) | FlowNet3D flow embedding · FlyingThings3D test | EPE: 0.2600 -> 0.2298; 11.6% error reduction | Helped |
| Replace 3D interpolation with set upconv refinement (ablation) | FlowNet3D · FlyingThings3D test | EPE: 0.2298 -> 0.1835; 20% | Helped |
| Add multiple inference resampling (ablation) | FlowNet3D with set upconv · FlyingThings3D test | EPE: 0.1835 -> 0.1694 | Helped |
| Add cycle-consistency regularization (ablation) | FlowNet3D with set upconv and resampling · FlyingThings3D test | EPE: 0.1694 -> 0.1626 | Helped |
| Fine-tune on 100 KITTI frames | FlowNet3D · KITTI scene flow (with ground points), test 50 frames | 3D EPE: 0.211 -> 0.144 | Helped |
| Use ICP-estimated flow for segmented ground points and FlowNet3D flow elsewhere | FlowNet3D + ICP · KITTI scene flow (with ground points), test 50 frames | 3D EPE: 0.211 -> 0.195 | Helped |
| Add scene flow before fitting a rigid motion | Point cloud registration · Point cloud warping registration | EPE: 0.384 -> 0.220 | Helped |
| Fit a rigid motion after applying scene flow | Point cloud registration · Point cloud warping registration | EPE: 0.220 -> 0.125 | Helped |
Closest papers by meaning
- FlowMamba: Learning Point Cloud Scene Flow with Global Motion Propagation (AAAI 2025 · similarity 0.88)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.88)
- 3DSFLabelling: Boosting 3D Scene Flow Estimation by Pseudo Auto-labelling (CVPR 2024 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.87)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.86)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.86)
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.