Losses

CycleLoss

class CycleLoss(loss_fn=<sphinx.ext.autodoc.importer._MockObject object>)

Bases: sphinx.ext.autodoc.importer._MockObject

Computes a cyclic loss between the original and reconstructed images of each domain

Parameters:loss_fn (optional) – the actual loss function to compute a pixelwise loss (the default is :class:`torch.nn.L1Loss`())

forward(target_a: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4f860>, target_b: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4f940>, rec_a: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4fc88>, rec_b: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4fcf8>)

Calculates the actual loss

Parameters:

• target_a (torch.Tensor) – the target image of domain A
• target_b (torch.Tensor) – the target image of domain B
• rec_a (torch.Tensor) – the reconstructed image of domain A
• rec_b (torch.Tensor) – the reconstructed image of domain B

Returns:

the loss value

Return type:

torch.Tensor

AdversarialLoss

class AdversarialLoss(loss_fn=<sphinx.ext.autodoc.importer._MockObject object>)

Bases: sphinx.ext.autodoc.importer._MockObject

Calculates an adversarial loss on the classification results of the fake images of both image domains (needed to update the generators)

Parameters:loss_fn (optional) – the actual loss function computing the losses (the default is :class:`torch.nn.BCELoss`())

forward(fake_a_cls: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4fe10>, fake_b_cls: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4ff98>)

Calculates the actual loss

Parameters:

• fake_a_cls (torch.Tensor) – classification result of the fake image in domain A
• fake_b_cls (torch.Tensor) – classification result of the fake image in domain B

Returns:

the loss value

Return type:

torch.Tensor

DiscriminatorLoss

class DiscriminatorLoss(loss_fn=<sphinx.ext.autodoc.importer._MockObject object>)

Bases: sphinx.ext.autodoc.importer._MockObject

Calculates a classical discriminator loss (classification whether image is real or fake)

Parameters:loss_fn (optional) – the actual loss function computing the losses (the default is :class:`torch.nn.BCELoss`())

forward(real_cl: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4fa20>, fake_cl: <sphinx.ext.autodoc.importer._MockObject object at 0x7f897ab4f8d0>)

Calculates the actual loss

Parameters:

• real_cl (torch.Tensor) – classification result of the real image
• fake_cl (torch.Tensor) – classification result of the fake image

Returns:

the loss value

Return type:

torch.Tensor