mlsanityData to 7 Oct 2026

Re-Evaluating LiDAR Scene Flow for Autonomous Driving

arXiv 2304.02150 ↗First posted 4 Apr 2023cs.CVcited by 17

Not on the map. Why?

Abstract (arXiv)

Popular benchmarks for self-supervised LiDAR scene flow (stereoKITTI, and FlyingThings3D) have unrealistic rates of dynamic motion, unrealistic correspondences, and unrealistic sampling patterns. As a result, progress on these benchmarks is misleading and may cause researchers to focus on the wrong problems. We evaluate a suite of top methods on a suite of real-world datasets (Argoverse 2.0, Waymo, and NuScenes) and report several conclusions. First, we find that performance on stereoKITTI is negatively correlated with performance on real-world data. Second, we find that one of this task's key components -- removing the dominant ego-motion -- is better solved by classic ICP than any tested method. Finally, we show that despite the emphasis placed on learning, most performance gains are caused by pre- and post-processing steps: piecewise-rigid refinement and ground removal. We demonstrate this through a baseline method that combines these processing steps with a learning-free test-time flow optimization. This baseline outperforms every evaluated method.

Key facts. Cited by 17 papers in the map (data through 7 Oct 2026). Builds on 43 of its 85 references that are in the map. First posted on arXiv 4 Apr 2023.

Builds on

43 of 85 references.

References · 43

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,223
  3. 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,485
  4. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,384
  5. 2019DeepSDF: Learning Continuous Signed Distance Functions for Shape Representationcited by 2,682
  6. 2020Implicit Neural Representations with Periodic Activation Functionscited by 2,214
  7. 2018Occupancy Networks: Learning 3D Reconstruction in Function Spacecited by 2,037
  8. 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,707
  9. 2018Learning Implicit Fields for Generative Shape Modelingcited by 1,060
  10. 2015A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimationcited by 878
  11. 2018PointCNN: Convolution On X-Transformed Pointscited by 764
  12. 2023Argoverse 2: Next Generation Datasets for Self-Driving Perception and Forecastingcited by 701
31 more
  1. 2019SAL: Sign Agnostic Learning of Shapes from Raw Datacited by 389
  2. 2018FlowNet3D: Learning Scene Flow in 3D Point Cloudscited by 241
  3. 2022KISS-ICP: In Defense of Point-to-Point ICP -- Simple, Accurate, and Robust Registration If Done the Right Waycited by 132
  4. 2020FLOT: Scene Flow on Point Clouds Guided by Optimal Transportcited by 129
  5. 2020RAFT-3D: Scene Flow using Rigid-Motion Embeddingscited by 121
  6. 2019MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequencescited by 117
  7. 2019HPLFlowNet: Hierarchical Permutohedral Lattice FlowNet for Scene Flow Estimation on Large-scale Point Cloudscited by 108
  8. 2019Just Go with the Flow: Self-Supervised Scene Flow Estimationcited by 98
  9. 2021Neural Scene Flow Priorcited by 91
  10. 2020PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Cloudscited by 74
  11. 2019Smooth Shells: Multi-Scale Shape Registration with Functional Mapscited by 72
  12. 2020FlowStep3D: Model Unrolling for Self-Supervised Scene Flow Estimationcited by 71
  13. 2020Self-Supervised Monocular Scene Flow Estimationcited by 64
  14. 2021Weakly Supervised Learning of Rigid 3D Scene Flowcited by 60
  15. 2022Motion Inspired Unsupervised Perception and Prediction in Autonomous Drivingcited by 59
  16. 2019Deep Rigid Instance Scene Flowcited by 54
  17. 2021Scalable Scene Flow from Point Clouds in the Real Worldcited by 49
  18. 2022Patchwork++: Fast and Robust Ground Segmentation Solving Partial Under-Segmentation Using 3D Point Cloudcited by 45
  19. 2020Scene Flow from Point Clouds with or without Learningcited by 44
  20. 2020Self-Supervised Learning of Non-Rigid Residual Flow and Ego-Motioncited by 41
  21. 2019SENSE: a Shared Encoder Network for Scene-flow Estimationcited by 38
  22. 2022Bi-PointFlowNet: Bidirectional Learning for Point Cloud Based Scene Flow Estimationcited by 38
  23. 2021HCRF-Flow: Scene Flow from Point Clouds with Continuous High-order CRFs and Position-aware Flow Embeddingcited by 36
  24. 2022What Matters for 3D Scene Flow Networkcited by 31
  25. 2022Deformation and Correspondence Aware Unsupervised Synthetic-to-Real Scene Flow Estimation for Point Cloudscited by 28
  26. 2019Mono-SF: Multi-View Geometry Meets Single-View Depth for Monocular Scene Flow Estimation of Dynamic Traffic Scenescited by 27
  27. 2020DeepLiDARFlow: A Deep Learning Architecture For Scene Flow Estimation Using Monocular Camera and Sparse LiDARcited by 24
  28. 2019PWOC-3D: Deep Occlusion-Aware End-to-End Scene Flow Estimationcited by 20
  29. 2018Motion-based Object Segmentation based on Dense RGB-D Scene Flowcited by 18
  30. 2022RMS-FlowNet: Efficient and Robust Multi-Scale Scene Flow Estimation for Large-Scale Point Cloudscited by 13
  31. 2020A Deep Temporal Fusion Framework for Scene Flow Using a Learnable Motion Model and Occlusionscited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 2; 2024: 4; 2025: 6; 2026: 5.

17 papers.

Sort:

  1. 2024SeFlow: A Self-Supervised Scene Flow Method in Autonomous Drivingcited by 38
  2. 2025Any4D: Unified Feed-Forward Metric 4D Reconstructioncited by 32
  3. 2023milliFlow: Scene Flow Estimation on mmWave Radar Point Cloud for Human Motion Sensingcited by 20
  4. 2023Multi-Body Neural Scene Flowcited by 12
  5. 2025SSF: Sparse Long-Range Scene Flow for Autonomous Drivingcited by 12
  6. 2025HiMo: High-Speed Objects Motion Compensation in Point Cloudscited by 11
  7. 2025DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Methodcited by 7
  8. 2026TeFlow: Enabling Multi-frame Supervision for Self-Supervised Feed-forward Scene Flow Estimationcited by 5
  9. 2025DoGFlow: Self-Supervised LiDAR Scene Flow via Cross-Modal Doppler Guidancecited by 4
  10. 2024Self-Supervised Multi-Frame Neural Scene Flowcited by 3
  11. 2024Let-It-Flow: Simultaneous Optimization of 3D Flow and Object Clusteringcited by 3
  12. 2026TerraSeg: Self-Supervised Ground Segmentation for Any LiDARcited by 2
5 more
  1. 2024Fast Kernel Scene Flowcited by 1
  2. 2026SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Datacited by 1
  3. 2025UniFlow: Zero-Shot LiDAR Scene Flow for Autonomous Vehiclescited by 0
  4. 2026Weakly Supervised Cross-Modal Learning for 4D Radar Scene Flow Estimationcited by 0
  5. 2026CorrelationFlow: A Training-Free Geometric Approach for LiDAR Scene Flow Estimationcited by 0