Deep Ensemble Learning for Human Action Recognition in Still Images

Table 9

Results for DELVS in the 1.5x cropped Willow action dataset.

Model	Strategy	Sensitivity for each class							Overall
		Inter.W.C.	Photog.	P.Music	R.Bike	R.Horse	Running	Walking	Acc
DELVS1	Hard	0.87	0.35	0.69	0.84	0.77	0.62	0.39	0.6351
	Soft	0.82	0.27	0.72	0.86	0.75	0.65	0.43	0.6414
	Tuning	0.82	0.3	0.73	0.86	0.74	0.68	0.43	0.6461
DELVS2	Hard	0.82	0.45	0.77	0.84	0.75	0.51	0.43	0.6509
	Soft	0.82	0.40	0.79	0.84	0.79	0.54	0.47	0.6619
	Tuning	0.85	0.29	0.76	0.86	0.86	0.51	0.61	0.6777
DELVS3	Hard	0.85	0.4	0.74	0.85	0.81	0.58	0.4	0.6493
	Soft	0.85	0.34	0.76	0.84	0.81	0.58	0.46	0.6556
	Tuning	0.85	0.4	0.79	0.84	0.79	0.59	0.49	0.6746