Automatic Aesthetics Evaluation of Robotic Dance Poses Based on Hierarchical Processing Network

Table 3

The performance comparison of different methods in comparative experiments.

Category	Method	Feature type	Type of input image	Fusion	Correct ratio (%)	Computation cost	Remarks
Conventional machine learning	Naïve Bayes	Hand-crafted features (Color + Shape + orientation)	RGB	Feature fusion	52.3	5.951	The approach in [10]
	Bayesian logistic regression				62.3	5.951
	RBF network				53.8	5.951
	AD tree				71.5	5.952
	Random forest				70.8	5.951
	Voted perceptron				71.5	5.954
	Bagging				64.6	5.951
	Rotation forest				70.8	5.955
	LWL				61.5	5.992
Deep learning	Convolution neural network	Automatically extracted features	HSV		69.2	0.020	Designed for comparison
			RGB		70.8	0.029
			Depth		71.5	0.017
			HSV + RGB + Depth	Input fusion	74.6	0.052
			HSV + RGB + Depth	Feature fusion	82.3	0.103	Our approach