List your company
ResearchFormerly WizWand

Video Recognition on Mini-Kinetics (test)

79.6AccuracyUpper
60.25665.27870.375.322Dec 29, 2020Apr 2, 2021Jul 6, 2021Oct 9, 2021Jan 12, 2022Apr 17, 2022Jul 21, 2022
Updated 8mo ago

Evaluation results

MethodLinks
UpperInput frames=16, GFLOPs=117.3, #3D=5.0, #Frame=16.02020.1279.6
Ada3DInput frames=16, GFLOPs=73.8, #3D=2.9, #Frame=11.82020.1279.2
UpperInput frames=8, GFLOPs=58.6, #3D=5.0, #Frame=8.02020.1279
Ada3DInput frames=8, GFLOPs=42.4, #3D=1.9, #Frame=6.92020.1278.9
Random FTInput frames=16, GFLOPs=78.0, #3D=2.9, #Frame=12.02020.1278.2
Random FTInput frames=8, GFLOPs=41.5, #3D=1.9, #Frame=6.82020.1277.4
RandomInput frames=16, GFLOPs=75.8, #3D=2.9, #Frame=11.82020.1275.2
RandomInput frames=8, GFLOPs=42.2, #3D=2.0, #Frame=6.92020.1274
NSNetBackbone=ResNet-50, GFLOPS=18.1G2022.0773.6
AdaFocusBackbone=ResNet-50, GFLOPS=38.6G2022.0772.9
FrameExitBackbone=ResNet-50, GFLOPS=19.7G2022.0772.8
Dynamic-STEBackbone=ResNet-50, GFLOPS=18.3G2022.0772.7
AdaFuseBackbone=ResNet-50, GFLOPS=23.0G2022.0772.3
VideoIQBackbone=ResNet-50, GFLOPS=20.4G2022.0772.3
AR-NetBackbone=ResNet-50, GFLOPS=32.0G2022.0771.7
SCSamplerBackbone=ResNet-50, GFLOPS=42.0G2022.0770.8
LiteEvalBackbone=ResNet-50, GFLOPS=99.0G2022.0761