Optical Flow Estimation on Sintel Final (test)
1.62EPEVideoFlow
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VideoFlowSupervision=Supervised, Multi-frame data usage=true2026.04 | 1.62 | — | — | — | — | |
| VideoFlow-MOFTemporal Context=Multi-frame2026.03 | 1.65 | — | — | — | 8.8 | |
| VideoFlow-BOFTemporal Context=Multi-frame2026.03 | 1.71 | — | — | — | 9.42 | |
| MemFlow-TTemporal Context=Multi-frame2026.03 | 1.83 | — | — | — | 9.83 | |
| StreamFlowTemporal Context=Multi-frame2026.03 | 1.87 | — | — | — | 10.68 | |
| GeoViTTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=true2025.12 | 1.88 | — | — | — | — | |
| GG-SSMmulti-frame inputs=true2024.12 | 1.9 | — | — | — | — | |
| FlowFormer++Training Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 1.94 | — | — | — | — | |
| DPFlowTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 1.97 | — | — | — | — | |
| WAFT-DINOv3-a2Temporal Context=Multi-frame2026.03 | 2.02 | — | — | — | 12.49 | |
| FlowDiffuserTemporal Context=Multi-frame2026.03 | 2.03 | — | — | — | 10.93 | |
| FlowDiffuserSupervision=Supervised2026.04 | 2.03 | — | — | — | — | |
| MemFlowmulti-frame inputs=true2024.12 | 2.08 | — | — | — | — | |
| SAM-FlowTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.08 | — | — | — | — | |
| SAMFlowTemporal Context=Multi-frame2026.03 | 2.08 | — | — | — | 11.28 | |
| FlowFormerTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.09 | — | — | — | — | |
| FlowFormerTemporal Context=Two-frame2026.03 | 2.09 | — | — | — | 11.67 | |
| FlowFormerSupervision=Supervised2026.04 | 2.09 | — | — | — | — | |
| GMFlow+Training Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.12 | — | — | — | — | |
| MatchAttentionEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.16 | — | — | — | — | |
| VideoFlow-BOFmulti-frame inputs=true2024.12 | 2.19 | — | — | — | — | |
| SKFlowTraining Data=C + T + S + K + H, Warm-start=true, Tiling technique=false2025.12 | 2.23 | — | — | — | — | |
| FlowFormer++multi-frame inputs=false2024.12 | 2.3 | — | — | — | — | |
| TransFlowmulti-frame inputs=true2024.12 | 2.33 | — | — | — | — | |
| FlowFormerEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.4 | — | — | — | — | |
| CRAFTTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.42 | — | — | — | — | |
| MegaFlowTemporal Context=Multi-frame2026.03 | 2.43 | — | — | — | 15.5 | |
| CrocoFlowTemporal Context=Two-frame2026.03 | 2.44 | — | — | — | 15.21 | |
| AnyFlowTemporal Context=Multi-frame2026.03 | 2.44 | — | — | — | 14.2 | |
| MatchFlowmulti-frame inputs=false2024.12 | 2.45 | — | — | — | — | |
| RPKNetEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.45 | — | — | — | — | |
| SKFlowmulti-frame inputs=false2024.12 | 2.46 | — | — | — | — | |
| SKFlowEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.46 | — | — | — | — | |
| GMATraining Data=C + T + S + K + H, Warm-start=true, Tiling technique=false2025.12 | 2.47 | — | — | — | — | |
| GMATemporal Context=Two-frame2026.03 | 2.47 | — | — | — | 13.5 | |
| GMFlowmulti-frame inputs=false2024.12 | 2.48 | — | — | — | — | |
| DDVMTemporal Context=Two-frame2026.03 | 2.48 | — | — | — | 16.55 | |
| EMD-LEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.55 | — | — | — | — | |
| VideoFlow-MOFmulti-frame inputs=true2024.12 | 2.56 | — | — | — | — | |
| SEA-RAFT (L)Temporal Context=Two-frame2026.03 | 2.6 | — | — | — | 15.71 | |
| GMFlowNetTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.65 | — | — | — | — | |
| RPKNetTemporal Context=Two-frame2026.03 | 2.66 | — | — | — | 16 | |
| S-FlowTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.67 | — | — | — | — | |
| RAFTmulti-frame inputs=false2024.12 | 2.71 | — | — | — | — | |
| GMFlowNetmulti-frame inputs=false2024.12 | 2.71 | — | — | — | — | |
| RAFTEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.71 | — | — | — | — | |
| GMFlowNetEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.71 | — | — | — | — | |
| GMAmulti-frame inputs=false2024.12 | 2.74 | — | — | — | — | |
| GMAEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.74 | — | — | — | — | |
| CRAFTEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.79 | — | — | — | — | |
| DIPEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 2.82 | — | — | — | — | |
| RAFT (warm-start)Training Data=C + T + S + K + H2020.03 | 2.86 | — | — | — | — | |
| RAFTTraining Data=C + T + S + K + H, Warm-start=true, Tiling technique=false2025.12 | 2.86 | — | — | — | — | |
| RAFTTemporal Context=Two-frame2026.03 | 2.86 | — | — | — | 16.37 | |
| RAFTSupervision=Supervised2026.04 | 2.86 | — | — | — | — | |
| GMATraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.88 | — | — | — | — | |
| GMFlowTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 2.9 | — | — | — | — | |
| RAFT (2-view)Training Data=C + T + S + K + H2020.03 | 3.18 | — | — | — | — | |
| RAFTTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 3.18 | — | — | — | — | |
| RAFT-ftLearning paradigm=Supervised, Fine-tuning=true2020.12 | 3.39 | — | — | — | — | |
| RAFT (2-view)Training Data=C + T + S/K2020.03 | 3.41 | — | — | — | — | |
| FM-RAFTTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 3.6 | — | — | — | — | |
| PWC-NetEvaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 3.93 | — | — | — | — | |
| SEA-RAFT(M)Evaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 4.04 | — | — | — | — | |
| ScopeFlowTraining Data=C + T + S/K2020.03 | 4.1 | — | — | — | — | |
| U2Flow (+FF)Supervision=Unsupervised, Multi-frame data usage=true, Bidirectional flow fusion (+FF)=true2026.04 | 4.1 | — | — | — | — | |
| SEA-RAFT(L)Evaluation Protocol=Zero-shot, Training Dataset=Scene Flow2025.10 | 4.11 | — | — | — | — | |
| U2FlowSupervision=Unsupervised2026.04 | 4.16 | — | — | — | — | |
| MaskFlowNetTraining Data=C + T + S + K + H2020.03 | 4.17 | — | — | — | — | |
| MaskFlowNetTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 4.17 | — | — | — | — | |
| SMURFSupervision=Unsupervised, Multi-frame data usage=true2026.04 | 4.18 | — | — | — | — | |
| SelFlow-ftSupervision Type=Supervised, Training Domain Condition (Labeled/Unlabeled/Related evaluation data)=Unlabeled evaluation data2020.06 | 4.26 | — | — | — | — | |
| VCN-ftSupervision Type=Supervised2020.06 | 4.4 | — | — | — | — | |
| VCNTraining Data=C + T + S + K + H2020.03 | 4.4 | — | — | — | — | |
| VCNTraining Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 4.4 | — | — | — | — | |
| IRR-PWCTraining Data=C + T + S/K2020.03 | 4.58 | — | — | — | — | |
| IRR-PWC+ftLearning paradigm=Supervised, Fine-tuning=true2020.12 | 4.58 | — | — | — | — | |
| PWC-Net+Training Data=C + T + S + K + H2020.03 | 4.6 | — | — | — | — | |
| PWC-Net+Training Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 4.6 | — | — | — | — | |
| PWC-Net+Supervision=Supervised2026.04 | 4.6 | — | — | — | — | |
| HD3Training Data=C + T + S/K2020.03 | 4.67 | — | — | — | — | |
| LiteFlowNet2Training Data=C + T + S + K + H2020.03 | 4.69 | — | — | — | — | |
| LiteFlowNet2Training Data=C + T + S + K + H, Warm-start=false, Tiling technique=false2025.12 | 4.69 | — | — | — | — | |
| M2FlowSupervision=Unsupervised, Multi-frame data usage=true2026.04 | 5.01 | — | — | — | — | |
| PWC-Netfine-tuning=true2018.02 | 5.04 | — | — | — | — | |
| PWC-NetTemporal Context=Two-frame2026.03 | 5.04 | — | — | — | 31.07 | |
| DCFlow2018.02 | 5.12 | — | — | — | — | |
| DCFlowTraining Data=S2020.03 | 5.12 | — | — | — | — | |
| PWC-Net-ftSupervision Type=Supervised2020.06 | 5.13 | — | — | — | — | |
| PWC-Net+ftLearning paradigm=Supervised, Fine-tuning=true2020.12 | 5.13 | — | — | — | — | |
| UnSAMFlowSupervision=Unsupervised, Semantic inputs=true2026.04 | 5.2 | — | — | — | — | |
| UPFlowLearning paradigm=Unsupervised2020.12 | 5.32 | — | — | — | — | |
| UPFlowSupervision=Unsupervised2026.04 | 5.32 | — | — | — | — | |
| FlowFields2018.02 | 5.36 | — | — | — | — | |
| MRFlow2018.02 | 5.38 | — | — | — | — | |
| LiteFlowNetfine-tuning=true2018.02 | 5.38 | — | — | — | — | |
| MRFlowTraining Data=S2020.03 | 5.38 | — | — | — | — | |
| FlowFields++2020.03 | 5.49 | — | — | — | — | |
| FlowNet2fine-tuning=true2018.02 | 5.74 | — | — | — | — | |
| FlowNet2-ftSupervision Type=Supervised2020.06 | 5.74 | — | — | — | — |