Video Action Recognition

• LSTM

• 3D ConvNets

image-20241022001842588

• large parameters

• extract feature

• Two-Stream Network

• 3D-Fused Two-Stream

• Inflated 3D(I3D)

  • infate to 3D from 2D
  • use parameters in 2D models
  • bootstrapping

• Temporal Segment Networks

  • segmental consensus
  • late fusion