Instructor Activity Recognition through Deep Spatiotemporal Features and Feedforward Extreme Learning Machines

Table 5

Performance comparison of various variants of ELM using a deep spatiotemporal representation of action motion templates of IAVID-1 dataset.

Method	IAVID (70-30 split)		MuHAVi-Uncut (LOCO)
	Accuracy	Computational Time	Accuracy	Computational Time
ELM	0.81250	6.05 m sec	82.04%	2280 sec
RELM [27]	0.81250	6.29 m sec	82.04%	2340 sec
SADE ELM [28]	0.235	0.9829 sec	50.98%	4080 sec
MCV-ELM [29]	0.1225	33.17sec	74.75%	5280 sec
MV-ELM [30]	0.1225	32.41 sec	74.75%	4380 sec