awesome-point-cloud-scene-flow

Awesome-Point-Cloud-Scene-Flow

- Recent papers (from 2019), the last three years are listed and previous papers are in the expandable list.
- welcome to add if any information is missing. 😎

Sorted by: whether open-sourced; the year of official publication; and the date of first public. Check the dataset section for scene flow dataset.

We encourage the community to open-source their code to support future research and foster collaboration 🚀. Check some codebases to start your work: KTH-RPL/OpenSceneFlow, kylevedder/SceneFlowZoo

---

2025

• [NeurIPS 25] DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Method [2508.17054][code]
• [TRO 25] (SeFlow++) HiMo: High-Speed Objects Motion Compensation in Point Cloud [2503.00803][code]
• [CVPR 25] VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow [2503.22328][code]
• [ICLR 25] EulerFlow: Neural Eulerian Scene Flow Fields [2410.02031][code]
• [RA-L 25] Flow4D: Leveraging 4D Voxel Network for LiDAR Scene Flow Estimation [2407.07995][code]
• [ICRA 25] SSF: Sparse Long-Range Scene Flow for Autonomous Driving [2501.17821][code]
• [arXiv 25] DoGFlow: Self-Supervised LiDAR Scene Flow via Cross-Modal Doppler Guidance [2508.18506][code]
• [arXiv 25] MambaFlow: A Novel and Flow-guided State Space Model for Scene Flow Estimation [2502.16907][code]

2024

• [NeurIPS 24] DiffSF: Diffusion Models for Scene Flow Estimation [2403.05327][code]
• [ECCV 24] SeFlow: A Self-Supervised Scene Flow Method in Autonomous Driving [2407.01702][code]
• [CVPR 24] 3DSFLabelling: Boosting 3D Scene Flow Estimation by Pseudo Auto-labelling [2402.18146] [code]
• [CVPR 24] ICP-Flow: LiDAR Scene Flow Estimation with ICP [2402.17351] [code]
• [CVPR 24] DifFlow3D: Toward Robust Uncertainty-Aware Scene Flow Estimation with Iterative Diffusion-Based Refinement [2311.17456][code]
• [ICRA 24] DeFlow: Decoder of Scene Flow Network in Autonomous Driving [2401.16122][code]
• [3DV 24] Multi-Body Neural Scene Flow [2310.10301][code]
• [TIV 24] Let-It-Flow: Simultaneous Optimization of 3D Flow and Object Clustering [2404.083636][code][1-Minute Video]
• [ICLR 24] ZeroFlow: Fast Zero Label Scene Flow via Distillation [2305.10424][code]
• [ECCV 24] I Can’t Believe It’s Not Scene Flow! [2403.04739]
• [WACV 24] OptFlow: Fast Optimization-Based Scene Flow Estimation Without Supervision [2401.02550]
• [WACV 24] Re-Evaluating LiDAR Scene Flow for Autonomous Driving [2304.02150]

2023

• [TPAMI 23] 3D Point-Voxel Correlation Fields for Scene Flow Estimation code
• [ICCV 23] DELFlow: Dense Efficient Learning of Scene Flow for Large-Scale Point Clouds [2308.04383][code]
• [ICCV 23] Fast Neural Scene Flow [2304.09121][code]
• [CVPR 23] Hidden Gems: 4D Radar Scene Flow Learning Using Cross-Modal Supervision [2303.00462][code]
• [CVPR 23] SCOOP: Self-Supervised Correspondence and Optimization-Based Scene Flow [2211.14020][code]
• [CVPR 23] Self-Supervised 3D Scene Flow Estimation Guided by Superpoints [2305.02528][code]
• [NeurIPS 23] GMSF: Global Matching Scene Flow [2305.17432] [code]
• [RAL 23] PT-FlowNet: Scene Flow Estimation on Point Clouds with Point Transformer
• [arXiv] Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Prior [2310.11284]
• [arXiv] ContrastMotion: Self-supervised Scene Motion Learning for Large-Scale LiDAR Point Clouds [2304.12589]
• [arXiv] GotFlow3D: Recurrent Graph Optimal Transport for Learning 3D Flow Motion in Particle Tracking [2210.17012]
• [arXiv] PointFlowHop: Green and Interpretable Scene Flow Estimation from Consecutive Point Clouds [2302.14193]
• [arXiv] Exploiting Implicit Rigidity Constraints via Weight-Sharing Aggregation for Scene Flow Estimation from Point Clouds [2303.02454]

2019 -2022

[Click me to expand]

2022

• [ECCV 22] FH-Net: A Fast Hierarchical Network for Scene Flow Estimation on Real-world Point Clouds code
• [ECCV 22] Dynamic 3D Scene Analysis by Point Cloud Accumulation [2207.12394] [code]
• [ECCV 22] What Matters for 3D Scene Flow Network [2207.09143] [code]
• [CVPR 22] RCP: Recurrent Closest Point for Scene Flow Estimation on 3D Point Clouds [2205.11028] [code]
• [CVPR 22] Deformation and Correspondence Aware Unsupervised Synthetic-to-Real Scene Flow Estimation for Point Clouds [2203.16895] [code]
• [IJCV] Learning Scene Dynamics from Point Cloud Sequences [2111.08755] [code]
• [RA-L&IROS 22] Self-Supervised Scene Flow Estimation with 4D Automotive Radar [2203.01137][code]
• [ACM MM 22] RPPformer-Flow: Relative Position Guided Point Transformer for Scene Flow Estimation [link] [code]
• [ECCV 22] Bi-PointFlowNet: Bidirectional Learning for Point Cloud Based Scene Flow Estimation [2207.07522] [code]
• [RAL 22] Estimation and Propagation: Scene Flow Prediction on Occluded Point Clouds
• [ICRA 22] RMS-FlowNet: Efficient and Robust Multi-Scale Scene Flow Estimation for Large-Scale Point Clouds [2204.00354]
• [AAAI 22] Self-Supervised Robust Scene Flow Estimation via the Alignment of Probability Density Functions [2203.12193]
• [arxiv] 3D Scene Flow Estimation on Pseudo-LiDAR: Bridging the Gap on Estimating Point Motion [2209.13130]
• [arXiv] PointConvFormer: Revenge of the Point-based Convolution [2208.02879]
• [arXiv] Learning Scene Flow in 3D Point Clouds with Noisy Pseudo Labels [2203.12655]

2021

• [CVPR 21] Self-Supervised Pillar Motion Learning for Autonomous Driving [2104.08683][code]
• [CVPR 21] Learning to Segment Rigid Motions from Two Frames [2101.03694][code]
• [CVPR 21 Oral] Weakly Supervised Learning of Rigid 3D Scene Flow [2102.08945][code]
• [CVPR 21] FESTA: Flow Estimation via Spatial-Temporal Attention for Scene Point Clouds [2104.00798] [code]
• [CVPRW 21] Occlusion Guided Scene Flow Estimation on 3D Point Clouds [2104.00798] [code]
• [RA-L 21] Scalable Scene Flow from Point Clouds in the Real World [2103.01306], Unofficial implementation code: kylevedder/zeroflow, Jabb0/FastFlow3D
• [3DV 21] Occlusion Guided Self-supervised Scene Flow Estimation on 3D Point Clouds [2104.04724]
• [AAAI 22] SCTN: Sparse Convolution-Transformer Network for Scene Flow Estimation [2105.04447]
• [CVPR 21] HCRF-Flow: Scene Flow from Point Clouds with Continuous High-order CRFs and Position-aware Flow Embedding [2105.07751]
• [CVPR 21 Oral] Self-Point-Flow: Self-Supervised Scene Flow Estimation from Point Clouds with Optimal Transport and Random Walk [2105.08248]
• [TIM 22] Residual 3D Scene Flow Learning with Context-Aware Feature Extraction [2109.04685]
• [NeurIPS 21] Accurate Point Cloud Registration with Robust Optimal Transport [2111.00648] [code]
• [NeurIPS 21 spotlight] Neural Scene Flow Prior [2111.01253] [code]

2020

• [ECCV 20] PointPWC-Net: A Coarse-to-Fine Network for Supervised and Self-Supervised Scene Flow Estimation on 3D Point Clouds [1911.12408][code]
• [ECCV 20] FLOT: Scene Flow on Point Clouds Guided by Optimal Transport [2007.11142][code]
• [3DV 20] Self-Supervised Learning of Non-Rigid Residual Flow and Ego-Motion [2009.10467] [code]
• [CVPR 20] Just Go With the Flow: Self-Supervised Scene Flow Estimation [1912.00497][code]
• [CVPR 21] PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Clouds [2012.00987] [code]
• VoxFlowNet: Learning Scene Flow in 3D Point Clouds through Voxel Grids [code]
• [3DV 20] Scene Flow from Point Clouds with or without Learning [2011.00320]
• [3DV 20] Adversarial Self-Supervised Scene Flow Estimation [2011.00551]
• [WACV 20] FlowNet3D++: Geometric Losses For Deep Scene Flow Estimation [1912.01438]
• [TIP 21] Hierarchical Attention Learning of Scene Flow in 3D Point Clouds [2010.05762]
• [CVPR 21] PWCLO-Net: Deep LiDAR Odometry in 3D Point Clouds Using Hierarchical Embedding Mask Optimization [2012.00972]
• [CVPR 21] FlowStep3D: Model Unrolling for Self-Supervised Scene Flow Estimation [2011.10147]
• [CVPR 21] RAFT-3D: Scene Flow using Rigid-Motion Embeddings [2012.00726]
• [IROS 20] PillarFlowNet: A Real-time Deep Multitask Network for LiDAR-based 3D Object Detection and Scene Flow Estimation [IROS20]

2019

• [ICCV 19] MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequences [1910.09165][code]
• [CVPR 19] FlowNet3D: Learning Scene Flow in 3D Point Clouds [1806.01411][code]
• [CVPR 19] HPLFlowNet: Hierarchical Permutohedral Lattice FlowNet for Scene Flow Estimation on Large-scale Point Clouds [1906.05332][code]

---

Dataset

• 2024-02-27: More and more datasets are available for scene flow estimation in autonomous driving (network input: 80k-107k points/frame). The following is a list of datasets that are commonly used in recent papers.

  • Argoverse 2.0 official website, processed available in DeFlow-OpenSF, ZeroFlow and av2 official
  • Waymo Open Dataset official website, processed available in SeFlow-OpenSF, ZeroFlow
  • nuScene official website, processed available in DeltaFlow-OpenSF
  • TruckScene official website, processed available in DoGFlow-OpenSF

• 2020-12-14: Since there is currently no raw dataset for Scene Flow Estimation with a point cloud as input (network input: max to 8,192 points/frame), the pioneers FlowNet3D and HPLFlowNet provide two versions of the dataset based on the raw dataset.

  • FlyingThings3D [official website]
  • KITTI 2015 [official website]
  • Processed by FlowNet3D [code]
  • Processed by HPLFlowNet [code]

  There are some differences in the way FlowNet3D and HPLFlowNet process data. [FlowNet3D only provides the code to process FlyingThings3D, HPLFlowNet provides code to process FlyingThings3D and KITTI15] Some papers will compare two kinds of data at the same time. [But at the moment there seems to be more comparisons on HPLFlowNet]

Public Leaderboard

• AV2 2024 Scene Flow Challenge, submission page: released first in CVPR 2024 Workshop
• Argoverse 2.0 Scene Flow , submission page: released first in CVPR 2023 Workshop