Table 1.
The architecture of TMD-Unet.
| Node | MSI | Encoder | DS | Decoder | Output |
|---|---|---|---|---|---|
| 1 | 16 × (3 × 3 Conv) |
[16 × (DC3 + ReLU)]2 + BN Max-pooling [16 × (DC2 + ReLU)]2 + BN Max-pooling [16 × (DC1 + ReLU)]2 + BN Max-pooling |
|
[16 × (DC3+ ReLU)]2 + BN Transposed Convolution [16 × (DC2+ ReLU)]2 + BN Transposed Convolution [16 × (DC1+ ReLU)]2 + BN Transposed Convolution |
|
| 2 | Max-pooling | [32 × (DC3 + ReLU)]2 + BN Max-pooling [32 × (DC2 + ReLU)]2 + BN Max-pooling [32 × (DC1 + ReLU)]2 + BN Max-pooling |
|
[32 × (DC3+ ReLU)]2 + BN Transposed Convolution [32 × (DC2+ ReLU)]2 + BN Transposed Convolution [32 × (DC1+ ReLU)]2 + BN Transposed Convolution |
|
| 3 | Max-pooling | [64 × (DC3 + ReLU)]2 + BN Max-pooling [64 × (DC2 + ReLU)]2 + BN Max-pooling [64 × (DC1 + ReLU)]2 + BN Max-pooling |
|
[64 × (DC3+ ReLU)]2 + BN Transposed Convolution [64 × (DC2+ ReLU)]2 + BN Transposed Convolution [64 × (DC1+ ReLU)]2 + BN Transposed Convolution |
|
| 4 | Max-pooling | [128 × (DC3 + ReLU)]2 + BN Max-pooling [128 × (DC2 + ReLU)]2 + BN Max-pooling [128 × DC1 + ReLU)]2 + BN Max-pooling |
|
[128 × (DC3+ ReLU)]2 + BN Transposed Convolution [128 × (DC2+ ReLU)]2 + BN Transposed Convolution [128 × (DC1+ ReLU)]2 + BN Transposed Convolution |
|
| Tr. | Max-pooling | [256 × (DC3 + ReLU)]2 + BN [256 × (DC2 + ReLU)]2 + BN [256 × (DC1 + ReLU)]2 + BN |
|||
DCx denotes the dilated convolution with a dilation rate equal to x. The superscript number indicates the number of repetitions. ‘Tr.’ is the transition.