Scene Flow Estimation on KITTI
0.021EPE (m)PointATA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PointATA2026.02 | 0.021 | 97.8 | 99.8 | 8.8 | — | |
| DiffFlow3DReference=ArXiv20232026.02 | 0.031 | 95.5 | 96.6 | 10.8 | — | |
| SPFlowNetSup.=Self.2023.05 | 0.0362 | 87.24 | 95.79 | 17.71 | — | |
| ICP-Flow2024.02 | 0.0423 | 94.3 | 94.42 | — | — | |
| MSBRNReference=ICCV20232026.02 | 0.044 | 4.4 | 95 | 20.8 | — | |
| FLOTSup.=Full.2023.05 | 0.056 | 75.5 | 90.8 | 24.2 | — | |
| RigidFlowSup.=Self.2023.05 | 0.0619 | 72.37 | 89.23 | 26.18 | — | |
| SCOOPReference=CVPR20232026.02 | 0.063 | 79.7 | 91 | 24.4 | — | |
| BPFReference=ECCV20222026.02 | 0.065 | 76.9 | 90.6 | 26.4 | — | |
| PointPWC-NetSup.=Full.2023.05 | 0.0694 | 72.81 | 88.84 | 26.48 | — | |
| 3DFlowReference=ECCV20222026.02 | 0.073 | 81.9 | 89 | 26.1 | — | |
| OGSFNetReference=CVPR20212026.02 | 0.075 | 70.6 | 86.9 | 32.7 | — | |
| RCPSup.=Self.2023.05 | 0.0763 | 78.56 | 89.21 | 18.49 | — | |
| PDF-FlowSup.=Self.2023.05 | 0.092 | 74.7 | 87 | 28.3 | — | |
| FESTAInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.0936 | 44.85 | 83.35 | — | — | |
| FESTAReference=CVPR20212026.02 | 0.097 | 44.9 | 83.3 | — | — | |
| FlowStep3DSup.=Self.2023.05 | 0.1021 | 70.8 | 83.94 | 24.56 | — | |
| PointPWC-NetInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.1094 | 35.98 | 73.84 | — | — | |
| FLOTReference=ECCV20202026.02 | 0.11 | 41.9 | 72.1 | 48.6 | — | |
| FLOTAttack scenario=Unattacked2024.04 | 0.11 | — | — | — | — | |
| Self-Point-FlowSup.=Self.2023.05 | 0.112 | 52.76 | 79.36 | 40.86 | — | |
| HPLFlowNetSup.=Full.2023.05 | 0.1169 | 47.83 | 77.76 | 41.03 | — | |
| HPLFlowNetAttack scenario=Unattacked2024.04 | 0.117 | — | — | — | — | |
| HPLFlowNetInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.119 | 30.83 | 64.76 | — | — | |
| RigidFlowTraining Dataset=KITTI2024.02 | 0.1192 | 40.99 | 69.77 | — | — | |
| SLIMSup.=Self.2023.05 | 0.1207 | 51.78 | 79.56 | 40.24 | — | |
| FlowNet3DInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.122 | 18.53 | 57.03 | — | — | |
| Just Go w/ FlowInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.122 | 25.37 | 57.85 | — | — | |
| HPLFlowNetAttack scenario=Attack on dimension 12024.04 | 0.136 | — | — | — | 0.162 | |
| HPLFlowNetAttack scenario=Attack on dimension 32024.04 | 0.137 | — | — | — | 0.171 | |
| HPLFlowNetAttack scenario=Attack on dimension 22024.04 | 0.156 | — | — | — | 0.333 | |
| FlowNet3DReference=CVPR20192026.02 | 0.169 | 25.4 | 57.9 | 78.9 | — | |
| FlowNet3DSup.=Full.2023.05 | 0.1767 | 37.38 | 66.77 | 52.71 | — | |
| HPLFlowNetAttack scenario=Attack on all dimensions2024.04 | 0.196 | — | — | — | 0.675 | |
| StableMTLStream Configuration=Multi-stream (Full)2025.06 | 0.2313 | — | — | — | — | |
| MeteorNetInput Modality=geo.-only, Generalization=zero-shot2021.04 | 0.251 | — | — | — | — | |
| FlowNet3D++Input Modality=geo.-only, Generalization=zero-shot2021.04 | 0.253 | — | — | — | — | |
| PoinPWC-NetSup.=Self.2023.05 | 0.2549 | 23.79 | 49.57 | 68.63 | — | |
| StableMTL-SStream Configuration=Single-stream2025.06 | 0.2618 | — | — | — | — | |
| Single-task baselineTraining Protocol=Single-task2025.06 | 0.2735 | — | — | — | — | |
| RigidFlowTraining Dataset=Waymo2024.02 | 0.2748 | 7.47 | 26.25 | — | — | |
| PSTNetInput=points, # Frames=32022.05 | 0.278 | — | — | — | — | |
| MeteorNetInput=points, # Frames=32022.05 | 0.282 | — | — | — | — | |
| FlowNet3DInput=points, # Frames=22022.05 | 0.287 | — | — | — | — | |
| FlowNet3DAttack scenario=Unattacked2024.04 | 0.299 | — | — | — | — | |
| DiffusionMTLFull Setting Adaptation=Yes2025.06 | 0.3502 | — | — | — | — | |
| Ego-MotionSup.=Self.2023.05 | 0.4154 | 22.09 | 37.21 | 80.96 | — | |
| Zero-MSFEvaluation Domain=In-distribution2025.01 | 0.452 | 39.8 | 44.3 | 87.3 | — | |
| Self-Mono-SF-kitti-trainEvaluation Domain=In-distribution, Training=Trained on KITTI, train split2025.01 | 0.454 | 34.5 | 43.5 | 85.3 | — | |
| Self-Mono-SF-kitti-eigenEvaluation Domain=In-distribution, Training=Trained on KITTI, eigen split2025.01 | 0.517 | 39.2 | 45.7 | 87 | — | |
| JTRFull Setting Adaptation=Yes2025.06 | 0.5176 | — | — | — | — | |
| Zero-MSF (exclude)Evaluation Domain=Out-of-distribution2025.01 | 0.641 | 39.2 | 43.1 | 89.9 | — | |
| FlowNet3DAttack scenario=Attack on dimension 12024.04 | 1.426 | — | — | — | 3.769 | |
| FLOTAttack scenario=Attack on dimension 32024.04 | 1.577 | — | — | — | 13.336 | |
| FlowNet3DAttack scenario=Attack on dimension 32024.04 | 1.588 | — | — | — | 4.311 | |
| FlowNet3DAttack scenario=Attack on dimension 22024.04 | 1.603 | — | — | — | 4.361 | |
| FLOTAttack scenario=Attack on dimension 12024.04 | 1.857 | — | — | — | 15.882 | |
| FLOTAttack scenario=Attack on dimension 22024.04 | 1.932 | — | — | — | 16.564 | |
| OpticalExpansion-drivingEvaluation Domain=Out-of-distribution, Training=Trained on Driving [63]2025.01 | 2.537 | 34.8 | 36.3 | 96.8 | — | |
| OpticalExpansion-kitti-trainvalEvaluation Domain=In-distribution, Training=Trained on Driving [63], and then finetune on KITTI, trainval split2025.01 | 2.682 | 35.9 | 36.7 | 96.7 | — | |
| OpticalExpansion-robustEvaluation Domain=Out-of-distribution, Training=Trained for the Robust Vision Challenge2025.01 | 2.702 | 35.9 | 36.6 | 96.9 | — | |
| OpticalExpansion-kitti-trainEvaluation Domain=In-distribution, Training=Trained on Driving [63], and then finetune on KITTI, train split2025.01 | 2.703 | 35.7 | 36.6 | 97 | — | |
| FlowNet3DAttack scenario=Attack on all dimensions2024.04 | 2.932 | — | — | — | 8.806 | |
| FLOTAttack scenario=Attack on all dimensions2024.04 | 3.159 | — | — | — | 27.718 |