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

Table 2

Comparison of verification accuracy between two-stream network and single-stream network.

Network	CS (top 1)	CV (top 1)
S-Stream	84.32%	91.94%
V-Stream	82.12%	92.85%
SV-GCN	85.51%	94.15%