Research map / 3D vision / Depth and stereo
SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Data
Qingwen Zhang, Xiaomeng Zhu, ChenHan Jiang, 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 |
|---|---|---|---|
| train only on synthetic SynFlow-4k (zero-shot) instead of 20% real labels | DeltaFlow · TruckScenes val | Dynamic Bucket-Normalized EPE: 0.402 -> 0.274 | Improved |
| train only on SynFlow-4k (zero-shot) vs 20% real labels | DeltaFlow · nuScenes val | Dynamic Bucket-Normalized EPE: 0.216 -> 0.242 | Hurt |
| pre-train on SynFlow-4k then fine-tune on 20% real labels | DeltaFlow · nuScenes val | Dynamic Bucket-Normalized EPE: 0.216 -> 0.157 | Improved |
| pre-train on SynFlow-4k then fine-tune on 20% real labels | DeltaFlow · TruckScenes val | Dynamic Bucket-Normalized EPE: 0.402 -> 0.266 | Improved |
| fine-tune SynFlow-4k prior on only 5% real labels (ablation) | DeltaFlow · nuScenes val | Dynamic Bucket-Normalized EPE: 0.216 -> 0.201 | Improved |
| increase fine-tuning labels from 5% to 20% (ablation) | SynFlow-4k + FT · nuScenes val | Dynamic Bucket-Normalized EPE: 0.201 -> 0.175 -> 0.157 | Improved |
| increase fine-tuning labels from 10% to 20% (ablation) | SynFlow-4k + FT · TruckScenes val | Dynamic Bucket-Normalized EPE: 0.261 -> 0.266 | No clear effect |
| scale synthetic data from 1k to 4k sequences (ablation) | DeltaFlow · Aeva zero-shot and TruckScenes | Dynamic Bucket-Normalized EPE | Improved |
| add synthetic SynFlow-4k to UniFlow real-data training (ablation) | UniFlow · Aeva zero-shot | Dynamic Bucket-Normalized EPE: 0.303 -> 0.263 | Improved |
| add greedy topology-coverage route selection (Policy 1) (ablation) | SynFlow 1k split · Aeva zero-shot | Dynamic Bucket-Normalized EPE: 0.346 -> 0.330 | Improved |
| add highway towns for speed regime (Policy 2) (ablation) | SynFlow 1k split · Aeva zero-shot | Dynamic Bucket-Normalized EPE: 0.330 -> 0.319 | Improved |
| add highway towns (Policy 2) on FD error (ablation) | SynFlow 1k split · Aeva zero-shot | Three-way EPE FD (cm): 50.21 -> 48.17 | Improved |
Closest papers by meaning
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.92)
- 3DSFLabelling: Boosting 3D Scene Flow Estimation by Pseudo Auto-labelling (CVPR 2024 · similarity 0.89)
- Motion Cues from Image-based Point Tracking for LiDAR Scene Flow Estimation (NeurIPS 2026 · similarity 0.89)
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · similarity 0.89)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.88)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (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.
