Deep Ensemble Learning for Human Action Recognition in Still Images

Table 2

Results for NCNN in the Willow action dataset.

Algorithm	Sensitivity for each class							Overall
	Inter.W.C.	Photog.	P.Music	R.Bike	R.Horse	Running	Walking	Acc	Loss
SURF	0.03	0.01	0.01	0.01	0.02	0.02	0.01	0.1908	NA
BOF	0.55	0.43	0.42	0.48	0.40	0.26	0.13	0.3735	NA
PBOF	0.48	0.47	0.38	0.43	0.4	0.36	0.21	0.3795	NA
GIST	0.55	0.23	0.41	0.39	0.28	0.28	0.32	0.3434	NA
RF	0.28	0.39	0.35	0.44	0.36	0.28	0.07	0.3153	NA
GBM	0.45	0.28	0.33	0.41	0.26	0.40	0.17	0.3193	NA
Voting	0.52	0.33	0.39	0.38	0.30	0.31	0.21	0.3394	NA
VGG16	0.76	0.69	0.71	0.94	0.84	0.6	0.07	0.6486	1.1642
VGG16_NCNN	0.79	0.52	0.68	0.92	0.9	0.66	0.3	0.6647	1.0545
VGG19	0.9	0.41	0.83	0.93	0.86	0.47	0.32	0.6647	1.0172
VGG19_NCNN	0.83	0.51	0.72	0.86	0.88	0.67	0.23	0.6506	0.9145
ResNet50	0.93	0.39	0.7	0.89	0.78	0.41	0.51	0.6466	0.9598
ResNet50_NCNN	0.79	0.24	0.75	0.92	0.82	0.48	0.55	0.6506	0.9019