A Novel Parameter Initialization Technique Using RBM-NN for Human Action Recognition

Table 12

Computational complexity with respect to time for the various datasets.

Dataset	Classes	Resolution	Frames per second (fps)	Input video sample	Training set sample	Testing set sample	Testing accuracy (%)	Training accuracy (%)	Training time	Average epochs
Weizmann	10	180 × 144	25	91	60	31	94.1	96.3	53.05	60
CAVIAR	13	384 × 288	25	29	20	9	97.4	99.1	47.08	40
UCF SA	11	720 × 480	10	150	102	4	94.6	98.2	74.87	39
KTH	6	160 × 120	25	200	140	60	94.52	98.33	102.12	59
CASIA	8	320 × 240	25	250	190	60	95.7	98.75	146.12	75
i3DPost	13	1920 × 1080	15	104	60	44	93.19	97.09	51.91	40
JHMDB	12	320 × 240	25	316	224	92	88.71	90.83	203.72	45
UCF101	101	320 × 240	25	600	400	200	84.21	88.64	283.87	80
HMDB51	51	320 × 240	30	90	60	30	82.11	87.12	54.12	40