List your company

Video Classification on Something-something v1 (val)

54.7Top-1 AccVoV3D-L
18.09227.59637.146.604Dec 13, 2017Oct 27, 2018Sep 10, 2019Jul 24, 2020Jun 7, 2021Apr 21, 2022Mar 5, 2023

Evaluation results

MethodLinks
VoV3D-LPretrain=Kinetics, Frame=32, Param (M)=5.8, GFLOPS=24.3 x 62020.1254.782
VoV3D-MPretrain=Kinetics, Frame=32, Param (M)=3.2, GFLOPS=12.8 x 62020.1253.281.1
TSMRGB+FlowBackbone=ResNet-50, #Frame=16+16, #Param.=48.6M2018.1152.681.9
TEABackbone=ResNet-50, Pretrain=ImageNet, Frame=16, Param (M)=24.4, GFLOPS=70 x 302020.1252.381.9
TEABackbone=ResNet-50, Pretrain=ImageNet, Frame=8, Param (M)=24.4, GFLOPS=35 x 302020.1251.780.5
E3D-LBackbone=E3D-L, Pretrain=No pretrain, Resolution=16 x 312^2, GFLOPs=18.3, Testing Strategy=1x12023.0351.178.7
TSM+TPNBackbone=ResNet-50, Pretrain=ImageNet, Frame=82020.1250.7—
STMBackbone=ResNet-50, Pretrain=ImageNet, Frame=16, GFLOPS=67 x 302020.1250.780.4
VoV3D-LFrame=32, Param (M)=5.8, GFLOPS=24.3 x 62020.1250.778.8
VoV3D-MFrame=32, Param (M)=3.2, GFLOPS=12.8 x 62020.1249.878
TSMEnBackbone=ResNet-50, #Frame=24, FLOPs/Video=98G, #Param.=48.6M2018.1149.778.5
ECOEnLiteRGB+FlowBackbone=BNIncep+3D Res18, #Frame=92+92, #Param.=300M2018.1149.5—
VoV3D-LFrame=16, Param (M)=5.8, GFLOPS=12.1 x 62020.1249.578
X3D-LBackbone=X3D-L, Pretrain=No pretrain, Resolution=16 x 312^2, GFLOPs=18.4, Testing Strategy=1x12023.0349.477.9
E3D-MBackbone=E3D-M, Pretrain=No pretrain, Resolution=16 x 224^2, GFLOPs=4.7, Testing Strategy=1x12023.0349.478.1
ip-CSN-152Frame=32, Param (M)=29.7, GFLOPS=74.0 x 102020.1249.3—
MoViNet-A1Backbone=MoViNet-A1, Pretrain=No pretrain, Resolution=50 x 172^2, GFLOPs=6, Testing Strategy=1x12023.0349.377.1
I3D + RSTGBackbone=3D ResNet-502019.0449.278.8
VoV3D-MFrame=16, Param (M)=3.2, GFLOPS=6.4 x 62020.124978.2
TSMBackbone=ResNet-101, Pretrain=Kinetics, Frame=8, Param (M)=24.3, GFLOPS=65 x 62020.1248.777.2
S3D-GBackbone=Inception, Input Features=RGB2017.1248.278.7
S3D-GBackbone=3D Inception2019.0448.278.7
S3D-GBackbone=InceptionV1, Pretrain=ImageNet, Resolution=64 x 256^2, GFLOPs=71.4, Testing Strategy=1x12023.0348.278.7
TANetBackbone=ResNet50, Pretrain=ImageNet, Resolution=16 x 256^2, GFLOPs=66, Testing Strategy=1x12023.0347.677.7
S3DBackbone=Inception, Input Features=RGB2017.1247.378.1
X3D-MBackbone=X3D-M, Pretrain=No pretrain, Resolution=16 x 224^2, GFLOPs=4.7, Testing Strategy=1x12023.0347.376.6
TSMBackbone=ResNet-50, #Frame=16, FLOPs/Video=65G, #Param.=24.3M2018.1147.277.1
TSM_16FFLOPs=65G, Param.=24.3M, Latency=29.0ms, Thrput.=39.5V/s2018.1147.2—
TSMBackbone=ResNet-50, Pretrain=Kinetics, Frame=16, Param (M)=24.3, GFLOPS=33 x 62020.1247.277
TSMBackbone=ResNet50, Pretrain=ImageNet, Resolution=16 x 256^2, GFLOPs=65, Testing Strategy=1x12023.0347.277.1
X3D-LFrame=16, Param (M)=5.6, GFLOPS=12.0 x 62020.1247.176.5
E3D-SBackbone=E3D-S, Pretrain=No pretrain, Resolution=13 x 160^2, GFLOPs=1.9, Testing Strategy=1x12023.0347.175.6
MoViNet-A0Backbone=MoViNet-A0, Pretrain=No pretrain, Resolution=50 x 172^2, GFLOPs=2.7, Testing Strategy=1x12023.0346.975
ECOEnLiteBackbone=BNIncep+3D Res18, #Frame=92, FLOPs/Video=267G, #Param.=150M2018.1146.4—
X3D-MFrame=16, Param (M)=3.3, GFLOPS=6.1 x 62020.1246.475.3
NL I3D + Joint GCNbackbone=ResNet-502018.0646.176.8
Non-local I3D + GCNBackbone=3D ResNet-50+GCN, #Frame=32x2clip, FLOPs/Video=303G^2x2, #Param.=62.2M^22018.1146.176.8
NL I3D + Joint GCNBackbone=3D ResNet-502019.0446.176.8
NL-I3D+GCNBackbone=3D ResNet-50, Pretrain=Kinetics, Frame=32, GFLOPS=303 x 62020.1246.176.8
I3DBackbone=Inception, Input Features=RGB2017.1245.876.5
I3DBackbone=3D Inception2019.0445.876.5
TSMBackbone=ResNet-50, #Frame=8, FLOPs/Video=33G, #Param.=24.3M2018.1145.674.2
TSM_8FFLOPs=33G, Param.=24.3M, Latency=17.4ms, Thrput.=77.4V/s2018.1145.6—
I3D replaceFLOPs=48G, Param.=33.0M, Latency=28.0ms, Thrput.=37.9V/s2018.1144.9—
X3D-SBackbone=X3D-S, Pretrain=No pretrain, Resolution=13 x 160^2, GFLOPs=2, Testing Strategy=1x12023.0344.674.4
NL I3Dbackbone=ResNet-502018.0644.476
Non-local I3DBackbone=3D ResNet-50, #Frame=32x2clip, FLOPs/Video=168G^1x2, #Param.=35.3M2018.1144.476
NL I3DBackbone=3D ResNet-502019.0444.476
NL I3DBackbone=InceptionV1, Pretrain=ImageNet+K400, Resolution=64 x 256^2, GFLOPs=334, Testing Strategy=1x12023.0344.476
MFNet-C101Backbone=3D ResNet-1012019.0443.973.1
I3D + Joint GCNbackbone=ResNet-502018.0643.375.1
I3D + Spatial-Temporal GCNbackbone=ResNet-502018.0642.874.7
C2D + RSTGBackbone=2D ResNet-502019.0442.873.6
I3D + Similarity GCNbackbone=ResNet-502018.0642.774.6
ECO Lite-16FBackbone=2D Inc+3D Res-182019.0442.2—
Two-stream TRNRGB+FlowBackbone=BNInception, #Frame=8+8, #Param.=36.6M2018.1142—
I3Dbackbone=ResNet-502018.0641.672.2
I3DBackbone=3D ResNet-50, #Frame=32x2clip, FLOPs/Video=153G^1x2, #Param.=28.0M2018.1141.672.2
I3DFLOPs=306G, Param.=35.3M, Latency=165.3ms, Thrput.=6.1V/s2018.1141.6—
I3DBackbone=3D ResNet-502019.0441.672.2
I3DBackbone=InceptionV1, Pretrain=ImageNet+K400, Resolution=64 x 256^2, GFLOPs=306, Testing Strategy=1x12023.0341.672.2
ECOBackbone=BNIncep+3D Res18, #Frame=16, FLOPs/Video=64G, #Param.=47.5M2018.1141.4—
ECO_16FFLOPs=64G, Param.=47.5M, Latency=30.6ms, Thrput.=45.6V/s2018.1141.4—
MFNet-C50Backbone=3D ResNet-502019.0440.370.9
ECOBackbone=BNIncep+3D Res18, #Frame=8, FLOPs/Video=32G, #Param.=47.5M2018.1139.6—
TRN-Multiscale (our impl.)Backbone=ResNet-50, #Frame=8, FLOPs/Video=33G, #Param.=31.8M2018.1138.968.1
Multi-scale TRNBackbone=Inception, Input Features=RGB2017.1234.463.2
MultiScale TRNbackbone=Inception2018.0634.463.2
TRN-MultiscaleBackbone=BNInception, #Frame=8, FLOPs/Video=16G, #Param.=18.3M2018.1134.4—
TRNBackbone=2D Inception2019.0434.4—
C2DBackbone=2D ResNet-502019.0431.764.7
I2DBackbone=Inception, Input Features=RGB2017.1231.469
TSN (our impl.)Backbone=ResNet-50, #Frame=8, FLOPs/Video=33G, #Param.=24.3M2018.1119.746.6
TSNBackbone=BNInception, #Frame=8, FLOPs/Video=16G, #Param.=10.7M2018.1119.5—
TSNBackbone=ResNet50, Pretrain=ImageNet, Resolution=8 x 256^2, GFLOPs=16, Testing Strategy=1x12023.0319.5—