RGB-D Human Action Recognition of Deep Feature Enhancement and Fusion Using Two-Stream ConvNet

Figure 6

Data sample layer, convs represents dilated convolution, followed by BN layer and ReLU layer.