Research map / Point clouds / Point cloud classification
Learning Optical Flow and Scene Flow With Bidirectional Camera-LiDAR Fusion
Haisong Liu, Tao Lu, Yihui Xu, Jia Liu, Limin Wang
Read the paperCheck 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 |
|---|---|---|---|
| add 3D-to-2D fusion to unidirectional 2D-to-3D fusion (ablation) | CamLiRAFT · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.056 -> 0.053 | Improved |
| add fusion stages S_F, S_C, S_D versus no fusion (ablation) | CamLiPWC · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.108 -> 0.070 | Improved |
| also fuse GRU hidden state (S_H) (ablation) | CamLiRAFT · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.053 -> 0.065 | Hurt |
| add learnable ScoreNet to nearest-neighbor interpolation (ablation) | CamLiPWC · FlyingThings3D val (ablation, half epochs) | EPE_2D: 2.60 -> 2.52 | Improved |
| use k=3 neighbors with MEAN/MAX instead of k=1 (ablation) | CamLiPWC · FlyingThings3D val (ablation, half epochs) | EPE_2D: 2.52 -> 2.54 | No clear effect |
| add learnable interpolation to selective fusion (ablation) | CamLiRAFT · KITTI train (generalization from FlyingThings3D) | EPE_2D: 3.64 -> 3.03 | Improved |
| selective fusion instead of concatenation (ablation) | CamLiPWC · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.075 -> 0.070 | Improved |
| gated fusion instead of addition (ablation) | CamLiPWC · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.074 -> 0.075 | No clear effect |
| selective fusion instead of concatenation, generalization (ablation) | CamLiRAFT · KITTI train (generalization from FlyingThings3D) | EPE_2D: 3.25 -> 3.03 | Improved |
| add inverse depth scaling (ablation) | CamLiPWC-L · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.123 -> 0.116 | Improved |
| add inverse depth scaling (ablation) | CamLiRAFT-L · FlyingThings3D val (ablation, half epochs) | EPE_3D: 0.105 -> 0.096 | Improved |
| detach gradient between branches (ablation) | CamLiPWC · FlyingThings3D (val subset) | 3D accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlowNet3D — FlowNet3D: Learning Scene Flow in 3D Point Clouds (CVPR 2019) · 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
- CamLiFlow: Bidirectional Camera-LiDAR Fusion for Joint Optical Flow and Scene Flow Estimation (CVPR 2022 · similarity 0.96)
- SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Data (ECCV 2026 · similarity 0.91)
- RPEFlow: Multimodal Fusion of RGB-PointCloud-Event for Joint Optical Flow and Scene Flow Estimation (ICCV 2023 · similarity 0.89)
- x^2-Fusion: Cross-Modality and Cross-Dimension Flow Estimation in Event Edge Space (CVPR 2026 · similarity 0.89)
- Bring Event into RGB and LiDAR: Hierarchical Visual-Motion Fusion for Scene Flow (CVPR 2024 · similarity 0.88)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.88)
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.
