Fig. 4. The schematic representation for CNN architectures.

a CNN_2 and b CNN_3 architectures, which consisted of two and three convolution/max pooling layers, respectively. Three ensuing fully connected common layers followed. The number of filters, the kernel size, the pooling size, the stride, and the padding scheme are also given. The padding type was the “SAME” for all CNNs. Other deeper architectures such as CNN_4F, CNN_5F, and CNN_6F are schematically described in Supplementary Fig. 8.