Research map / 3D vision / Depth and stereo
Fast Neural Scene Flow
Xueqian Li, Jianqiao Zheng, Francesco Ferroni, Jhony Kaesemodel Pontes, Simon Lucey
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 |
|---|---|---|---|
| replace Chamfer distance loss with distance transform (DT) loss | neural scene flow prior · Waymo Open (full point cloud) | total time: 18.39 s -> 0.58 s | Improved |
| replace Chamfer distance loss with DT loss | neural scene flow prior · Waymo Open (full point cloud) | Acc5: 74.16 -> 84.73 | Improved |
| replace Chamfer distance loss with DT loss | neural scene flow prior · Argoverse (full point cloud) | 3D end-point error: 0.078 -> 0.071 | Mixed |
| replace 8-layer MLP with complex PE linear model (Chamfer loss) (ablation) | NSFP baseline · Waymo Open (full point cloud) | total time: 18.39 s -> 10.20 s | Improved |
| replace deep network with complex PE linear model (DT loss) (ablation) | DT-based method · Waymo Open (full point cloud) | Acc5: 84.73 -> 71.27 | Hurt |
| replace deep network with linear model (DT loss) (ablation) | DT-based method · Waymo Open (full point cloud) | total time: 0.58 s -> 0.49 s | No clear effect |
| use runtime optimization instead of pretrained learning methods on OOD lidar | Ours (8,192 pts) · Waymo Open (8,192 points) | Acc5: 2.46 -> 77.53 (vs FLOT) | Improved |
| use DT grid cell larger than 0.1 m | DT loss · Argoverse 2D scene | accuracy | Hurt |
| replace Chamfer loss in NSFP++ with DT loss | NSFP++ · Argoverse with ego-motion compensation | time (ms): 16188 -> 2001 | Improved |
| compare our method with NSFP++ with ego-motion compensation | Ours · Argoverse with ego-motion compensation | Acc10: 62.61 -> 46.82 | Hurt |
Closest papers by meaning
- Neural Scene Flow Prior (NeurIPS 2021 · similarity 0.92)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.90)
- Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within Unified Neural Representations (CVPR 2025 · similarity 0.88)
- LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes (CVPR 2024 · similarity 0.87)
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · similarity 0.87)
- STGC-NeRF: Spatial-Temporal Geometric Consistency for LiDAR Neural Radiance Fields in Dynamic Scenes (AAAI 2025 · 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.
