Table 7 Dual metaheuristic GAN architecture specifications.
Component | Layer type | Input size | Output size | Activation | Parameters | Identity block |
|---|---|---|---|---|---|---|
Generator | ||||||
Encoder Block 1 | Conv2D + BatchNorm | 256 × 256 × 3 | 128 × 128 × 64 | LeakyReLU | 12,352 | GIB-1 |
Encoder Block 2 | Conv2D + BatchNorm | 128 × 128 × 64 | 64 × 64 × 128 | LeakyReLU | 73,856 | GIB-2 |
Encoder Block 3 | Conv2D + BatchNorm | 64 × 64 × 128 | 32 × 32 × 256 | LeakyReLU | 295,168 | GIB-3 |
Encoder Block 4 | Conv2D + BatchNorm | 32 × 32 × 256 | 16 × 16 × 512 | LeakyReLU | 1,180,160 | GIB-4 |
Bottleneck | Conv2D + BatchNorm | 16 × 16 × 512 | 8 × 8 × 1024 | LeakyReLU | 4,719,616 | - |
Decoder Block 1 | ConvTranspose2D + BatchNorm | 8 × 8 × 1024 | 16 × 16 × 512 | ReLU | 4,719,104 | GIB-5 |
Decoder Block 2 | ConvTranspose2D + BatchNorm | 16 × 16 × 1024 | 32 × 32 × 256 | ReLU | 2,359,808 | GIB-6 |
Decoder Block 3 | ConvTranspose2D + BatchNorm | 32 × 32 × 512 | 64 × 64 × 128 | ReLU | 589,952 | GIB-7 |
Decoder Block 4 | ConvTranspose2D + BatchNorm | 64 × 64 × 256 | 128 × 128 × 64 | ReLU | 147,520 | GIB-8 |
Output layer | ConvTranspose2D | 128 × 128 × 128 | 256 × 256 × 3 | Tanh | 6,147 | – |
Discriminator | ||||||
Input block | Conv2D + LeakyReLU | 256 × 256 × 3 | 128 × 128 × 64 | LeakyReLU | 12,352 | DIB-1 |
Block 1 | Conv2D + BatchNorm | 128 × 128 × 64 | 64 × 64 × 128 | LeakyReLU | 73,856 | DIB-2 |
Block 2 | Conv2D + BatchNorm | 64 × 64 × 128 | 32 × 32 × 256 | LeakyReLU | 295,168 | DIB-3 |
Block 3 | Conv2D + BatchNorm | 32 × 32 × 256 | 16 × 16 × 512 | LeakyReLU | 1,180,160 | DIB-4 |
Block 4 | Conv2D + BatchNorm | 16 × 16 × 512 | 8 × 8 × 1024 | LeakyReLU | 4,719,616 | DIB-5 |
Output block | Conv2D | 8 × 8 × 1024 | 4 × 4 × 1 | Sigmoid | 16,385 | – |
