List your company

Optical Flow Estimation on MPI Sintel Final (train)

1.17Endpoint Error (EPE)HD3F-ft
1.06721.76112.4553.1489Dec 6, 2016Jul 2, 2018Jan 26, 2020Aug 22, 2021Mar 18, 2023Oct 11, 2024May 8, 2026

Evaluation results

MethodLinks
HD3F-ftFine-tuned=true, Time (s)=0.082018.121.17———
SelFlowLearning Paradigm=Supervised, Fine-tuned=true, Multi-frame=true, Number of Parameters=4.79 M2020.031.77———
LiteFlowNet-ftCategory=Lightweight CNN, fine-tuned=true2018.051.78———
LiteFlowNet-arXivTime (s)=0.09, fine-tuned=true2018.091.78———
LiteFlowNetfine-tuning=true2018.021.78———
LiteFlowNetParameters=5.37M2019.041.78———
LiteFlowNetFine-tuning=true2020.061.78———
LiteFlowNet-ftFine-tuning status=Yes, Params (M)=5.37, FLOPS (G)=163.5, Time (s) 1080Ti=0.055, Time (s) TX2=0.9072021.031.78———
FDFlowNetFine-tuning=true2020.061.93———
FlowNet2-ft-sintelbackbone=GPU, runtime=123ms, improved_option=true, fine-tuned=Sintel2016.122.01———
FlowNet2-ftTime (s)=0.12, fine-tuned=true2018.092.01———
FlowNet2fine-tuning=true2018.022.01———
FlowNet2-ftFine-tuned=true, Time (s)=0.122018.122.01———
FlowNet2Fine-tuning=true2020.062.01———
FlowNet2-ftFine-tuning status=Yes, Params (M)=162.52, FLOPS (G)=24836.4, Time (s) 1080Ti=0.116, Time (s) TX2=1.5472021.032.01———
MemFlow-TExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.06———
MemFlowExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.08———
PWC-Net-ft-finalTime (s)=0.03, fine-tuned=true, protocol=more weight to the final pass2018.092.08———
PWC-Netfine-tuning=true2018.022.08———
PWC-Net-ftFine-tuned=true, Time (s)=0.032018.122.08———
PWC-Net-ft-finalParameters=8.75M2019.042.08———
PWC-NetFine-tuning=true2020.062.08———
PWC-NetLearning Paradigm=Supervised, Fine-tuned=true, Number of Parameters=8.75 M2020.032.08———
PWC-Net-ftFine-tuning status=Yes, Params (M)=8.75, FLOPS (G)=90.8, Time (s) 1080Ti=0.034, Time (s) TX2=0.4852021.032.08———
StreamFlowExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.11———
VideoFlow-BOFExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.19———
FlowNet2-ft-sintelCategory=Heavyweight CNN, fine-tuned=Sintel2018.052.19———
PWC-Net-ftTime (s)=0.03, fine-tuned=true2018.092.21———
LiteFlowNet-CVPRTime (s)=0.09, fine-tuned=true2018.092.23———
LiteFlowNet-ftFine-tuned=true, Time (s)=0.092018.122.23———
LiteFlowNetLearning Paradigm=Supervised, Fine-tuned=true, Number of Parameters=5.37 M2020.032.23———
PWC-Net_ROBTime (s)=0.03, fine-tuned=true, data=Sintel, KITTI 2015, HD1K2018.092.29———
PWC-NetTraining Set=Sintel2020.032.31———
FlowNetC+Time (s)=0.05, fine-tuned=false2018.092.34———
PWC-Net+Time (s)=0.03, fine-tuned=true, protocol=training protocol II2018.092.34———
PWC-Net+Parameters=8.75M2019.042.34———
PWC-Net-small-ftTime (s)=0.02, fine-tuned=true2018.092.45———
PWC-NetFine-tuning=true, Model Variant=small2020.062.45———
GMFlow#refine=12026.052.480.512.8115.76
FlowSeek (T)#refine=42026.052.480.432.6316.6
IRR-PWCParameters=6.36M2019.042.51———
IRR-PWCLearning Paradigm=Supervised, Fine-tuned=true, Number of Parameters=6.36 M2020.032.51———
VideoFlow-MOFExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.56———
SMURFheavier backbone=true, multi-frame=true, trained on validation set=true2020.112.58———
SegFlow+ftfine-tuned=true2017.092.61———
Devonfine-tuning=true2018.022.67———
UPFlowheavier backbone=true, trained on validation set=true2020.112.67———
MEMFOFExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.7———
RAFTExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.71———
FastFlowNet-ftFine-tuning status=Yes, Params (M)=1.37, FLOPS (G)=12.2, Time (s) 1080Ti=0.011, Time (s) TX2=0.1762021.032.71———
RAFT#refine=322026.052.710.512.9817.62
MeFlowExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.062.75———
FlowNetS+Time (s)=0.01, fine-tuned=true2018.092.76———
Ours#refine=02026.052.810.643.3616.99
GMFlow#refine=02026.052.960.723.4517.7
FlowFieldsCategory=Conventional2018.053.06———
FlowFieldsbackbone=CPU, runtime=22,810ms, improved_option=true2016.123.06———
FlowNet2backbone=GPU, runtime=123ms, improved_option=true2016.123.14———
FlowNet2Time (s)=0.12, fine-tuned=false2018.093.14———
FlowNet2fine-tuning=false2018.023.14———
FlowNet2Fine-tuned=false, Time (s)=0.122018.123.14———
FlowNet2Parameters=162.5 M2019.043.14———
FlowNet2Fine-tuning=false2020.063.14———
FlowNet2Time (s)=0.122020.033.14———
FlowNet2Fine-tuning status=No, Params (M)=162.52, FLOPS (G)=24836.4, Time (s) 1080Ti=0.116, Time (s) TX2=1.5472021.033.14———
MDFlowTraining Dataset=Sintel2022.113.14———
FlowNet2-CSS-ft-sdbackbone=GPU, runtime=69ms, improved_option=true2016.123.17———
FlowNet2-CSSbackbone=GPU, runtime=69ms, improved_option=true2016.123.23———
Flow1DExtra data=None, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D2025.063.27———
UFlowTraining Dataset=Sintel2022.113.39———
UFlow-traintrained on validation set=true2020.113.39———
FlowNet2-css-ft-sdbackbone=GPU, runtime=31ms, improved_option=true2016.123.5———
FlowNet2Category=Heavyweight CNN2018.053.54———
FlowNet2-cssbackbone=GPU, runtime=31ms, improved_option=true2016.123.54———
EpicFlowCategory=Conventional2018.053.56———
EpicFlow+backbone=CPU, runtime=42,600ms, improved_option=true2016.123.56———
DeepFlowCategory=Conventional2018.053.57———
DeepFlow+backbone=CPU, runtime=51,940ms, improved_option=true2016.123.57———
SimFlowTraining Dataset=Sintel2022.113.57———
SimFlowtrained on validation set=true2020.113.57———
MR-Flow2017.053.59———
MRFlowTime (s)=480, fine-tuned=false2018.093.59———
MRFlowTime (s)=4802018.123.59———
MR-FlowParameters=N/A2019.043.59———
STFlowheavier backbone=true, trained on validation set=true2020.113.59———
FullFlowTime (s)=240, fine-tuned=false2018.093.6———
FullFlowTime (s)=2402018.123.6———
MaskFlownetTime (s)=0.062020.033.61———
UnrolledCosttrained on validation set=true2020.113.61———
VCNTime (s)=0.182020.033.62———
ARFlow-MVLearning Paradigm=Unsupervised, Multi-frame=true, Number of Parameters=2.37 M2020.033.69———
ARFlow-MVmulti-frame=true, trained on validation set=true2020.113.69———
EpicFlow2017.093.7———
Devonfine-tuning=false2018.023.72———
MaskFlownet-STime (s)=0.032020.033.72———
ARFlowTraining Set=Sintel2020.033.73———
ARFlowLearning Paradigm=Unsupervised, Number of Parameters=2.24 M2020.033.73———
ARFlowTraining Dataset=Sintel2022.113.73———
ARFlowtrained on validation set=true2020.113.73———
SEA-RAFT (S)Extra data=TartanAir, Protocol=Zero-shot, Training=FlyingChairs + FlyingThings3D + TartanAir2025.063.74———