Skip to main content
. 2021 Feb 20;11(2):359. doi: 10.3390/diagnostics11020359

Table 4.

The general structure of MobileNet CNN.

Layer Label Input Layer Dimension Filter and Stride Size
Convolution/S2 224 × 224 × 3 Filter size = 3 × 3 × 3 × 32
Stride = 2
Convolution/dw8/S1 112 × 112 × 32 Filter size = 3 × 3 × 32 dw
Stride = 1
Convolution/S1 112 × 112 × 32 Filter size 1 × 1 × 32 ×64
Stride = 1
Convolution/dw/S2 112 × 112 × 64 Filter size = 3 × 3 × 64 dw
Stride = 2
Convolution/S1 56 × 56 × 64 Filter size 1 × 1 × 64 × 128
Stride = 1
Convolution/dw/S1 56 × 56 × 128 Filter size = 3 × 3 × 128 dw
Stride = 2
Convolution/S1 56 × 56 × 128 Filter size 1 × 1 × 128 × 128
Stride = 1
Convolution/dw/S2 56 × 56 × 128 Filter size = 3 × 3 × 128 dw
Stride = 2
Convolution/S1 28 × 28 × 128 Filter size 1 × 1 × 128 × 256
Stride = 1
Convolution/dw/S1 28 × 28 × 256 Filter size = 3 × 3 × 256 dw
Stride = 1
Convolution/S1 28 × 28 × 256 Filter size 1 × 1 × 256 × 256
Stride = 1
Convolution/dw/S2 56 × 56 × 128 Filter size = 3 × 3 × 256 dw
Stride = 2
Convolution/S1 14 × 14 × 256 Filter size 1 × 1 × 256 × 512
Stride = 1
5 × Convolution dw/S1
5 × Convolution S1
14 × 14 × 512
14 × 14 × 512
Filter size = 3 × 3 × 512 dw
Filter size 1 × 1 × 512 × 512
Stride = 1
Convolution dw/S2 14 × 14 × 512 Filter size = 3 × 3 × 512 dw
Stride 2
Convolution/S1 7 × 7 × 512 Filter size 1 × 1 × 512 × 512
Stride = 1
Convolution dw/S2 7 × 7 × 1024 Filter size = 3 × 3 × 1024 dw
Stride 2
Convolution/S1 7 × 7 × 1024 Filter size 1 × 1 × 1024 × 1024
Stride = 1
Pooling Average Pooling= 7 × 7
Stride = 1
1 × 1 × 1024
FC Layer 1 ×1 × 1000

dw stands for depthwise.