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

Table 4

Comparison of verification accuracy between st-gcn network and improved Nonlocal-stgcn.

Network	CS (top 1)	CV (top 1)
Nonlocal-stgcn	84.35%	91.94%
St-gcn	81.5%	88.76%