Visual Computing for Industry, Biomedicine, and Art

Table 2 The ablation studies of different modules in STTG-Net, reporting results for the joint angle errors on Human3.6M. “√” indicates that the module is used in experiment, “×” means that the part is removed from experiment

From: STTG-net: a Spatio-temporal network for human motion prediction based on transformer and graph convolution network

	T-Transformer	S-GCN	TPE	PR	MAE
	T-Transformer	S-GCN	TPE	PR	80 ms	160 ms	320 ms	400 ms
(1)	√	×	×	×	0.59	0.83	1.16	1.28
(2)	×	√	×	×	0.49	0.69	0.98	1.09
(3)	√	√	×	×	0.28	0.54	0.88	1.00
(4)	√	√	√	×	0.27	0.52	0.86	0.98
(5)	√	√	√	√	0.28	0.50	0.80	0.89

The best results are presented in bold

Back to article page