Table 2 Model configuration for each method.
Parameters | CycleGAN | CycleGAN with Attention UNET & Resnet | SAN (ours) |
|---|---|---|---|
Image Input (pixels) | 256 × 256 × 3 | 256 × 256 × 3 | 256 × 256 × 3 |
Batch Size | 1 | 1 | 1 |
Epochs | 10 | 6 | 5 |
Generator | Vanilla | Attention-UNET | Attention-UNET |
Discriminator | Patch-GAN | Resnet-18 | Resnet-18 |
Learning Rate | \(\:1\bullet\:{10}^{-4}\) | Generator: \(\:1\bullet\:{10}^{-4}\) Discriminator: \(\:1\bullet\:{10}^{-5}\) | Generator: \(\:1\bullet\:{10}^{-3}\) Discriminator: \(\:1\bullet\:{10}^{-4}\) |
\(\:{\varvec{\lambda\:}}_{\varvec{G}\varvec{A}\varvec{N}}\) | 1 | 1 | 1 |
\(\:{\varvec{\lambda\:}}_{\varvec{c}\varvec{y}\varvec{c}\varvec{l}\varvec{e}}\) | 10 | 10 | 100 |
\(\:{\varvec{\lambda\:}}_{\varvec{s}\varvec{e}\varvec{g}}\) | 0 | 0 | 1.0 |
