Suppose the current input image size is known convolution layer (W), the size of the field of view may be a convolutional neurons (F.), Using a step size (S), the number (P) of the edge padding a zero value, to obtain the size of the output characteristics of FIG. calculated as: [(W-F + 2P) / S] +1
Pooling layer can gradually reduce the size of the space expressed, thereby reducing the number of parameters and computational overhead of the network; the same cell layer can function to control the action of over-fitting.