Table 3.

Detailed structure of CNN architecture. Conv. is used for convolutional layers, Max-Pool. is used for max pooling layers, and FC is used for fully connected layers.

Layer No.	Type	Filter Size	Stride	# Filters	FC Units	Input
Layer 1	Conv.	3 × 3	1 × 1	64	-	33 × 33
Layer 2	Conv.	3 × 3	1 × 1	64	-	64 × 33 × 33
Layer 3	Conv.	3 × 3	1 × 1	64	-	64 × 33 × 33
Layer 4	Maxpool.	3 × 3	2 × 2	-	-	64 × 33 × 33
Layer 5	Conv.	3 × 3	1 × 1	128	-	64 × 16 × 16
Layer 6	Conv.	3 × 3	1 × 1	128	-	128 × 16 × 16
Layer 7	Conv.	3 × 3	1 × 1	128	-	128 × 16 × 16
Layer 8	Maxpool.	3 × 3	2 × 2	-	-	128 × 16 × 16
Layer 9	FC	-	-	-	512	6272
Layer 10	FC	-	-	-	256	512
Layer 11	FC	-	-	-	2	256