An Action Recognition Method for Volleyball Players Using Deep Learning

Table 1

The data format changes after convolution and pooling.

Database input	Convolution (13 × 13)	Pooling (4 × 4)	Convolution (8 × 8)	Pooling (4 × 4)	Fully connected	Fully connected
Florence3D action: (@180 × 180)	46@168 × 168	46@42 × 42	78@35 × 35	78@9 × 9	128@1 × 1	9
MSR action pairs: (3@260 × 260)	46@248 × 248	46@62 × 62	78@55 × 55	78@14 × 14	128@1 × 1	12
UTKinect action: (3@208 × 208)	46@196 × 196	46@49 × 49	78@42 × 42	78@11 × 11	128@1 × 1	10