Table 1.
List of the different hyperparameters optimized over.
| Hyperparameters | Values |
|---|---|
| Differentiable augmentation [34] | TRUE/FALSE |
| Activation fn of discriminator | ReLU/LeakyRelu/Elu/Selu |
| Activation fn of generator | ReLU/LeakyRelu/Elu/Selu |
| Normalization layer of discriminator | BatchNorm [35]/InstanceNorm [36] |
| Normalization layer of generator | BatchNorm [35]/InstanceNorm [36] |
| Number of filters of discriminator | 16/32/64/128 |
| Number of filters of generator | 16/32/64/128 |
| Use spectral norm for discriminator | TRUE/FALSE |
| Use spectral norm for generator | TRUE/FALSE |
| Weight initialization function | Normal/Xavier/Xavier Uniform/Kaiming He |
| Weight initialization gain | 0.01/0.02/0.1/1.0 |
| Gradient penalty loss weight (WGAN-GP only) | 0/0.1/1.0/10.0 |
| Weight clipping value (WGAN only) | 0/0.01/0.1 |
| Feature matching loss weight | 0/1.0/10.0 |
| VGG loss weight | 0/1.0 /10.0 |
| Learning rate | 0.00004/0.00005/0.0001/0.0002/0.001 |
| Use of label smoothing [29] | TRUE/FALSE |
| Use of data augmentation | TRUE/FALSE |