Table A1.
Channel, Kernel, Stride, Padding, and Activation denote the configurations of the convolution layers in the blocks.
| Network | Block | Count | Channel | Kernel | Stride | Padding | Activation |
|---|---|---|---|---|---|---|---|
| Generator | encoder | 1 | 128 | 4 × 4 | 2 | same | LRelu |
| 2 | 256 | 1 | LRelu + BN | ||||
| 3 | 512 | ||||||
| 4 | 1024 | ||||||
| 5 | 1024 | ||||||
| 6 | 1024 | ||||||
| 7 | 1024 | ||||||
| decoder | 1 | 1024 | 4 × 4 | 2 | same | LRelu + BN | |
| 2 | 1024 + 1024 # | ||||||
| 3 | 1024 + 1024 # | ||||||
| 4 | 1024 + 1024 # | ||||||
| 5 | 512 + 512 # | ||||||
| 6 | 256 + 256 # | ||||||
| 7 | 128 + 128 # | ||||||
| 1 | - | - | - | Tanh | |||
| Discriminator | 1 | 128 | 4 × 4 | 2 | 1 | LRelu | |
| 2 | 256 | LRelu + BN | |||||
| 3 | 512 | ||||||
| 4 | 1024 | ||||||
| 5 | 1 | 1 | 0 | Sigmoid |
BN: Batch Normalization; LRelu: Leaky Rectified Linear Unit; Tanh: Hyperbolic Tangent function; #: U-net connect.