kaira.losses.adversarial.WassersteinGANLoss

Inheritance diagram for WassersteinGANLoss

class kaira.losses.adversarial.WassersteinGANLoss

Bases: BaseLoss

Wasserstein GAN Loss Module.

This module implements the WGAN loss from Arjovsky et al. 2017.

Methods

__init__	Initialize the WassersteinGANLoss module.
forward	Forward pass through the WassersteinGANLoss module.
forward_discriminator	Forward pass for discriminator.
forward_generator	Forward pass for generator.

__init__()

Initialize the WassersteinGANLoss module.

forward_discriminator(real_pred: Tensor, fake_pred: Tensor) → Tensor

Forward pass for discriminator.

Parameters:

• real_pred (torch.Tensor) – Discriminator outputs for real data.

• fake_pred (torch.Tensor) – Discriminator outputs for fake data.

Returns:

Discriminator loss.

Return type:

torch.Tensor

forward_generator(fake_pred: Tensor) → Tensor

Forward pass for generator.

Parameters:

fake_pred (torch.Tensor) – Discriminator outputs for fake data.

Returns:

Generator loss.

Return type:

torch.Tensor

forward(pred: Tensor, is_real: bool, for_discriminator: bool = True) → Tensor

Forward pass through the WassersteinGANLoss module.

Parameters:

• pred (torch.Tensor) – Discriminator outputs.

• is_real (bool) – Whether predictions are for real data.

• for_discriminator (bool) – Whether calculating loss for discriminator. Default is True.

Returns:

The Wasserstein loss.

Return type:

torch.Tensor