Research map / Point clouds / Point cloud classification
SeFlow: A Self-Supervised Scene Flow Method in Autonomous Driving
Qingwen Zhang, Yi Yang, Peizheng Li, Olov Andersson, Patric Jensfelt
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 dynamic Chamfer loss (L_dcham) to Chamfer loss (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.0962 -> 0.0916 | Improved |
| add static loss (L_static) (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.0916 -> 0.0779 | Improved |
| add static loss (L_static) (ablation) | SeFlow · Argoverse 2 val | EPE FD: 0.181 -> 0.220 | Hurt |
| add cluster-consistency classification loss with upper-bound flow (L_dcls) (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.0779 -> 0.0643 | Improved |
| keep Chamfer loss alongside the three proposed losses (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.0651 -> 0.0643 | No clear effect |
| use upper-bound object flow for cluster consistency instead of average or max (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.078 (avg) -> 0.064 | Improved |
| use SeFlow static/dynamic loss instead of FastFlow3D or DeFlow style weighting (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.094 -> 0.078 | Improved |
| train SeFlow with fewer data (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: SeFlow 20% 0.078 vs ZeroFlow 200% 0.076 and 100% 0.088 | Improved |
| swap DeFlow backbone for FastFlow3D backbone (ablation) | SeFlow · Argoverse 2 val | EPE 3-way: 0.059 -> 0.065 | Mixed |
Later papers that compared with this method
1 later accepted paper reports SeFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Motion Cues from Image-based Point Tracking for LiDAR Scene Flow Estimation (NeurIPS 2026 · better on 6 of 6 shared results)
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
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.91)
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · similarity 0.90)
- Motion Cues from Image-based Point Tracking for LiDAR Scene Flow Estimation (NeurIPS 2026 · similarity 0.90)
- 3DSFLabelling: Boosting 3D Scene Flow Estimation by Pseudo Auto-labelling (CVPR 2024 · similarity 0.89)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.89)
- SplatFlow: Self-Supervised Dynamic Gaussian Splatting in Neural Motion Flow Field for Autonomous Driving (CVPR 2025 · 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.
