Skip to main content
. 2020 Sep 9;11:558126. doi: 10.3389/fpls.2020.558126

Table 3.

Detailed structure of C-DenseNet variants.

Layers Output Size C-DenseNet-121(k=32) C-DenseNet-169(k=32) C-DenseNet-201(k=32)
Convolution 640 × 640 7 × 7 conv,stride2
Pooling 320 × 320 3 × 3max pool,stride2
Dense Block 1 320 × 320
(1×1 conv3×3 conv)×6
Transition Layer 1 320 × 320 1 × 1 conv
160 × 160 2 × 2 average pool
CBAM Layer 1 160 × 160 scale × 1
Dense Block 2 160 × 160
(1×1 conv3×3 conv)×12
Transition Layer 2 160 × 160 1 × 1 conv
80 × 80 2 × 2 average pool
CBAM Layer 2 80 × 80 scale × 1
Dense Block 3 80 × 80
(1×1 conv3×3 conv)×24
(1×1 conv3×3 conv)×32
(1×1 conv3×3 conv)×48
Transition Layer3 80 × 80 1 × 1 conv
40 × 40 2 × 2 average pool
CBAM Layer 3 40 × 40 1 × 1scale
Dense Block 4 40 × 40
(1×1 conv3×3 conv)×16
(1×1 conv3×3 conv)×32
(1×1 conv3×3 conv)×32
Classification Layer 1 × 1 40 × 40 average pooling
fully-connected, softmax