Research map / Point clouds / Point cloud classification
GenFlow3D: Generative Scene Flow Estimation and Prediction on Point Cloud Sequences
Hanlin Li, Wenming Weng, Yueyi Zhang, Zhiwei Xiong
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 multi-scale diffusion refinement (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.1796 -> 0.0646 | Improved |
| use transformer-based denoiser instead of setconv (ablation) | GenFlow3D · Argoverse 2 | EPE3D SF1/SF2: 8.3% for SF1 and 7.3% for SF2 lower EPE3D | Improved |
| add hidden state to condition signal (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.3337 -> 0.0646 | Improved |
| add correlation feature to condition signal (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.6008 -> 0.0646 | Improved |
| add encoder feature to condition signal (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.0805 -> 0.0699 | Improved |
| add coarse flow information to condition signal (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.0723 -> 0.0699 | Improved |
| add temporal initialization from previous full flow (ablation) | GenFlow3D · Argoverse 2 | EPE3D SF3: 11.1% drop on EPE3D for predicting SF3 | Improved |
| initialize from deepest-level instead of full previous flow (ablation) | GenFlow3D · Argoverse 2 | Overall EPE3D: 0.0646 -> 0.0689 | Hurt |
| use sequence history (SF2 vs SF1) | GenFlow3D · nuScenes | EPE3D: 0.0162 -> 0.0139 (14.2% lower) | Improved |
| lengthen input sequence from N=2 to N=4 (ablation) | GenFlow3D · nuScenes | EPE3D SF4: 0.0175 -> 0.0152 | Improved |
| increase prediction length from M=1 to M=2 (ablation) | GenFlow3D · nuScenes (N=3) | EPE3D SF4: 0.0161 -> 0.0172 | Hurt |
| use diffusion decoder vs DifFlow3D baseline | GenFlow3D · nuScenes SF2 | EPE3D: 0.0221 -> 0.0139 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PV-RAFT — PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Clouds (CVPR 2021) · better on 2 of 2 shared results
- PointPWC-Net — PointPWC-Net: Cost Volume on Point Clouds for (Self-)Supervised Scene Flow Estimation (ECCV 2020) · better on 2 of 2 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
- Learning Scene Dynamics from Point Cloud Sequences (IJCV 2022 · similarity 0.91)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · similarity 0.90)
- DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Method (NeurIPS 2025 · similarity 0.88)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.88)
- FlowMamba: Learning Point Cloud Scene Flow with Global Motion Propagation (AAAI 2025 · similarity 0.88)
- DiffSF: Diffusion Models for Scene Flow Estimation (NeurIPS 2024 · 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.
