Table 1 Architectural details of the proposed DermoNet
From: DermoNet: densely linked convolutional neural network for efficient skin lesion segmentation
Input image, 384 ×512×3 | |||||
|---|---|---|---|---|---|
Encoder | Decoder | ||||
Input | Filter | Output | Input | Filter | Output |
i-1 | 7 ×7, /2 | 192 ×256×64 | d-4-1 | 1 ×1 | 6 ×8×128 |
i-2 | 3 ×3, /2 | 96 ×128×64 | d-4-2 | 3 ×3, *2 | 12 ×16×128 |
e-1-1 | 3 ×3, /2 | 48 ×64×64 | d-4-3 | 1 ×1 | 12 ×16×256 |
e-1-2 | 3 ×3 | 48 ×64×64 | d-3-1 | 1 ×1 | 12 ×16×64 |
e-1-3 | 3 ×3 | 48 ×64×64 | d-3-2 | 3 ×3, *2 | 24 ×32×64 |
e-1-4 | 3 ×3 | 48 ×64×64 | d-3-3 | 1 ×1 | 24 ×32×128 |
e-2-1 | 3 ×3, /2 | 24 ×32×128 | d-2-1 | 1 ×1 | 24 ×32×32 |
e-2-2 | 3 ×3 | 24 ×32×128 | d-2-2 | 3 ×3, *2 | 48 ×64×32 |
e-2-3 | 3 ×3 | 24 ×32×128 | d-2-3 | 1 ×1 | 48 ×64×64 |
e-2-4 | 3 ×3 | 24 ×32×128 | d-1-1 | 1 ×1 | 48 ×64×16 |
e-3-1 | 3 ×3, /2 | 12 ×16×256 | d-1-2 | 3 ×3, *2 | 96 ×128×16 |
e-3-2 | 3 ×3 | 12 ×16×256 | d-1-3 | 1 ×1 | 96 ×128×64 |
e-3-3 | 3 ×3 | 12 ×16×256 | o-1 | 3 ×3, *2 | 192 ×256×32 |
e-3-4 | 3 ×3 | 12 ×16×256 | o-2 | 3 ×3 | 192 ×256×32 |
e-4-1 | 3 ×3, /2 | 6 ×8×512 | o-3 | 2 ×2, *2 | 384 ×512×1 |
e-4-2 | 3 ×3 | 6 ×8×512 | |||
e-4-3 | 3 ×3 | 6 ×8×512 | |||
e-4-4 | 3 ×3 | 6 ×8×512 | |||
Output image, 384 ×512×1 | |||||