Table 2.

IrisDenseNet connectivity and output feature map size of each dense block (Conv, BN, and ReLU represent convolutional layer, batch normalization layer, and rectified linear unit layer, respectively. Cat, B-Conv, and Pool indicate concatenation layer, bottleneck convolution layer, and pooling layer, respectively) (Here, dense blocks 1 and 2 have the same number of convolution layers, and dense blocks 3, 4, and 5 have the same number of convolution layers) (Convolutional layers with “*” mean that these layers include BN and ReLU. Transition layers are a combination of max-pooling and B-Conv)

Block	Name/Size	No. of Filters	Output Feature Map Size (Width × Height × Number of Channel)
Dense Block-1	Conv-1_1*/3 × 3 × 3	64	300 × 400 × 64
	Conv-1_2*/3 × 3 × 64	64
	Cat-1	-	300 × 400 × 128
Transition layer-1	B-Conv-1/1 × 1	64	300 × 400 × 64
	Pool-1/2 × 2	-	150 × 200 × 64
Dense Block-2	Conv-2_1*/3 × 3 × 64	128	150 × 200 × 128
	Conv-2_2*/3 × 3 × 128	128
	Cat-2	-	150 × 200 × 256
Transition layer-2	B-Conv-2/1 × 1	128	150 × 200 × 128
	Pool-2/2 × 2	-	75 × 100 × 128
Dense Block-3	Conv-3_1*/3 × 3 × 128	256	75 × 100 × 256
	Conv-3_2*/3 × 3 × 256	256
	Cat-3	-	75 × 100 × 512
	Conv-3_3*/3 × 3 × 256	256	75 × 100 × 256
	Cat-4	-	75 × 100 × 768
Transition layer-3	B-Conv-3/1 × 1	256	75 × 100 × 256
	Pool-3/2 × 2	-	37 ×50 × 256
Dense Block-4	Conv-4_1*/3 × 3 × 256	512	37 ×50 × 512
	Conv-4_2*/3 × 3 × 512	512
	Cat-5	-	37 ×50 × 1024
	Conv-4_3*/3 × 3 × 512	512	37 ×50 × 512
	Cat-6	-	37 ×50 × 1536
Transition layer-4	B-Conv-4/1 × 1	512	37 ×50 × 512
	Pool-4/2 × 2	-	18 × 25 × 512
Dense Block-5	Conv-5_1*/3 × 3 × 512	512	18 × 25 × 512
	Conv-5_2*/3 × 3 × 512	512
	Cat-7	-	18 × 25 × 1024
	Conv-5_3*/3 × 3 × 512	512	18 × 25 × 512
	Cat-8	-	18 × 25 × 1536
Transition layer-5	B-Conv-5/1 × 1	512	18 × 25 × 512
	Pool-5/2 × 2	-	9 × 12 × 512