Table 1.

The CNN architecture for kernel classification. The Conv, FC, and Avg pool stand for convolutional layer, fully connected layer, and average pooling layer, respectively.

Type/Stride	Filter Size	Number of Filters	Output Size
Conv/s1	$3\times 3$	32	$30\times 30\times 32$
Conv/s1	$3\times 3$	32	$28\times 28\times 32$
Avg pool/s2	$2\times 2$	-	$14\times 14\times 32$
Conv/s1	$3\times 3$	64	$12\times 12\times 64$
Conv/s1	$3\times 3$	64	$10\times 10\times 64$
Conv/s1	$3\times 3$	64	$8\times 8\times 64$
Avg pool/s1	$7\times 7$	-	$2\times 2\times 64$
FC-256
FC-128
Sigmoid