Skip to main content

Table 2 DenseNet-161 adapted structure

From: Does imbalance in chest X-ray datasets produce biased deep learning approaches for COVID-19 screening?

Layers

Output size

DenseNet-161

Convolution

112 x 112

Conv. 7 x 7, stride 2

Pooling

56 x 56

Max pool 3 x 3, stride 2

Dense block (1)

56 x 56

[1×1 conv. 3×3 conv.] x 6

Transition layer (1)

56 x 56

Conv. 1 x 1

 

28 x 28

2 x 2 average pool, stride 2

Dense block (2)

28 x 28

[1×1 conv. 3×3 conv.] x 12

Transition layer (2)

28 x 28

Conv. 1 x 1

 

14 x 14

2 x 2 average pool, stride 2

Dense block (3)

14 x 14

[1×1 conv. 3×3 conv.] x 36

Transition layer (3)

14 x 14

Conv. 1 x 1

 

7 x 7

2 x 2 average pool, stride 2

Dense block (4)

7 x 7

[1×1 conv. 3×3 conv.] x 24

Classification layer

1 x 1

7 x 7 global average pool

  

2D fully-connected, softmax