Optical Flow on MVSEC 1.0 (indoor_flying1)
0.11EPEEDCFlow (+in2)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EDCFlow (+in2)dt=1, Input=E, Learning paradigm=SL, training_inclusion=flying22025.06 | 0.11 | 2 | |
| ECN*dt=1, training_split=80%, evaluation_split=20%2020.03 | 0.21 | 0.01 | |
| ScaleEventBackbone=ViT-S/16, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.29 | 0.03 | |
| STPBackbone=Swin-T/7, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.31 | 0.03 | |
| ECDDPBackbone=Swin-T/7, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.36 | 0.04 | |
| EV-MGRFlowNetdt=1, Input=E, Learning paradigm=USL2025.06 | 0.41 | 17 | |
| MultiCM (Burgers')dt=1, Input=E, Learning paradigm=MB2025.06 | 0.42 | 10 | |
| EDCFlow (+in3)dt=1, Input=E, Learning paradigm=SL, training_inclusion=flying32025.06 | 0.42 | 1 | |
| ECDDPBackbone=ViT-S/16, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.51 | 0.11 | |
| Brebion et al.dt=1, Input=E, Learning paradigm=MB2025.06 | 0.52 | 10 | |
| Oursdt=12020.03 | 0.56 | 1 | |
| STE-FlowNetdt=1, Input=E, Learning paradigm=SSL2025.06 | 0.57 | 10 | |
| Zhu et al.dt=1, Input=E, Learning paradigm=USL2025.06 | 0.58 | 0 | |
| EVFlow+dt=12020.03 | 0.58 | 0 | |
| STPBackbone=ViT-S/16, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.58 | 0.05 | |
| ECDPBackbone=ResNet-50, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.6 | 0.35 | |
| ECDPBackbone=ViT-S/16, Protocol=Event Pretraining + Fully-Supervised2026.03 | 0.61 | 0.05 | |
| DCEIFlowdt=1, Input=E+I1, Learning paradigm=SL2025.06 | 0.75 | 155 | |
| DCEFlowProtocol=RGB initialization + Fully-Supervised2026.03 | 0.75 | 0.6 | |
| Voxel GridMeasurement=polarity, Kernel=trilinear2019.04 | 0.96 | 1.47 | |
| ESTMeasurement=time stamps, Kernel=exponential2019.04 | 0.96 | 1.27 | |
| Gehrigdt=12020.03 | 0.96 | 0.91 | |
| EDCFlowdt=1, Input=E, Learning paradigm=SL2025.06 | 0.97 | 219 | |
| ESTMeasurement=time stamps, Kernel=learnt2019.04 | 0.97 | 0.91 | |
| Voxel GridMeasurement=time stamps, Kernel=trilinear2019.04 | 0.98 | 1.2 | |
| ESTMeasurement=time stamps, Kernel=trilinear2019.04 | 1 | 1.35 | |
| EV-FlowNetdt=1, Input=E, Learning paradigm=SSL2025.06 | 1.03 | 220 | |
| EV-FlowNetMeasurement=count, Kernel=trilinear2019.04 | 1.03 | 2.2 | |
| ESTMeasurement=time stamps, Kernel=alpha2019.04 | 1.03 | 1.34 | |
| EVFlowdt=12020.03 | 1.03 | 2.2 | |
| IDNet-4dt=1, Input=E, Learning paradigm=SL, re-implemented=true2025.06 | 1.04 | 186 | |
| TMAdt=1, Input=E, Learning paradigm=SL2025.06 | 1.06 | 363 | |
| E-RAFTdt=1, Input=E, Learning paradigm=SL2025.06 | 1.1 | 572 | |
| Zerosdt=12020.03 | 1.1 | 1 | |
| IDNet-8dt=1, Input=E, Learning paradigm=SL, re-implemented=true2025.06 | 1.13 | 294 | |
| Event FrameMeasurement=time stamps, Kernel=trilinear2019.04 | 1.17 | 2.44 | |
| Two-Channel ImageMeasurement=time stamps, Kernel=trilinear2019.04 | 1.17 | 1.5 | |
| Two-Channel ImageMeasurement=count, Kernel=trilinear2019.04 | 1.21 | 4.49 | |
| ESTBackbone=ResNet-18, Protocol=RGB initialization + Fully-Supervised2026.03 | 1.24 | 5.09 | |
| EDCFlow (+in2)dt=4, Input=E, Learning paradigm=SL, training_inclusion=flying22025.06 | 1.28 | 524 | |
| EDCFlow (+in3)dt=4, Input=E, Learning paradigm=SL, training_inclusion=flying32025.06 | 1.3 | 556 | |
| EV-MGRFlowNetdt=4, Input=E, Learning paradigm=USL2025.06 | 1.5 | 867 | |
| MultiCM (Burgers')dt=4, Input=E, Learning paradigm=MB2025.06 | 1.69 | 1,295 | |
| STE-FlowNetdt=4, Input=E, Learning paradigm=SSL2025.06 | 1.77 | 1,470 | |
| DCEIFlowdt=4, Input=E+I1, Learning paradigm=SL2025.06 | 2.08 | 2,147 | |
| Zhu et al.dt=4, Input=E, Learning paradigm=USL2025.06 | 2.18 | 2,420 | |
| EV-FlowNetdt=4, Input=E, Learning paradigm=SSL2025.06 | 2.25 | 2,470 | |
| TMAdt=4, Input=E, Learning paradigm=SL2025.06 | 2.43 | 2,991 | |
| EDCFlowdt=4, Input=E, Learning paradigm=SL2025.06 | 2.59 | 2,816 | |
| E-RAFTdt=4, Input=E, Learning paradigm=SL2025.06 | 2.81 | 4,025 | |
| IDNet-4dt=4, Input=E, Learning paradigm=SL, re-implemented=true2025.06 | 3.3 | 4,272 | |
| IDNet-8dt=4, Input=E, Learning paradigm=SL, re-implemented=true2025.06 | 4.31 | 5,895 |