3D Deep Heterogeneous Manifold Network for Behavior Recognition

Table 4

Performance comparison on the NTU-RGBD dataset using cross-subject and cross-view protocol.

Methods	Accuracy
	Xsub (%)	Xview (%)
Lie group [4]	50.1	82.8
Deep-LSTM [23]	60.7	67.3
ST-LSTM [29]	69.2	77.7
TCN [7]	74.3	83.1
GCA-LSTM [30]	74.4	82.8
HCN [31]	86.5	91.1
ST-GR [32]	86.9	92.3
ST-GCN [8]	81.5	88.3
DGNN [33]	87.5	94.3
Ours	85.3	93.8