import torch
import torch.nn as nn
import torch.nn.functional as F
from .loss import GeneratorLoss, DiscriminatorLoss
from ..utils import reduce
__all__ = ['AuxiliaryClassifierGeneratorLoss', 'AuxiliaryClassifierDiscriminatorLoss']
[docs]class AuxiliaryClassifierGeneratorLoss(GeneratorLoss):
r"""Auxiliary Classifier GAN (ACGAN) loss based on a from
`"Conditional Image Synthesis With Auxiliary Classifier GANs
by Odena et. al. " <https://arxiv.org/abs/1610.09585>`_ paper
Args:
reduction (str, optional): Specifies the reduction to apply to the output.
If ``none`` no reduction will be applied. If ``mean`` the outputs are averaged over batch size.
If ``sum`` the elements of the output are summed.
override_train_ops (function, optional): A function is passed to this argument,
if the default ``train_ops`` is not to be used.
"""
def forward(self, logits, labels):
return F.cross_entropy(logits, labels, reduction=self.reduction)
r"""Defines the standard ``train_ops`` used by the Auxiliary Classifier generator loss.
The ``standard optimization algorithm`` for the ``discriminator`` defined in this train_ops
is as follows (label_g and label_d both could be either real labels or generated labels):
1. :math:`fake = generator(noise, label_g)`
2. :math:`value_1 = classifier(fake, label_g)`
3. :math:`value_2 = classifier(real, label_d)`
4. :math:`loss = loss\_function(value_1, label_g) + loss\_function(value_2, label_d)`
5. Backpropagate by computing :math:`\nabla loss`
6. Run a step of the optimizer for discriminator
Args:
generator (torchgan.models.Generator): The model to be optimized. For ACGAN, it must require
labels for training
discriminator (torchgan.models.Discriminator): The discriminator which judges the
performance of the generator.
optimizer_discriminator (torch.optim.Optimizer): Optimizer which updates the ``parameters``
of the ``discriminator``.
real_inputs (torch.Tensor): The real data to be fed to the ``discriminator``.
device (torch.device): Device on which the ``generator`` and ``discriminator`` is present.
batch_size (int): Batch Size of the data infered from the ``DataLoader`` by the ``Trainer``.
labels (torch.Tensor, optional): Labels for the data.
Returns:
Scalar value of the loss.
"""
[docs] def train_ops(self, generator, discriminator, optimizer_generator, device, batch_size, labels=None):
if self.override_train_ops is not None:
return self.override_train_ops(generator, discriminator, optimizer_generator, device, batch_size, labels)
if generator.label_type == 'required' and labels is None:
raise Exception('GAN model requires label for training')
noise = torch.randn(batch_size, generator.encoding_dims, device=device)
optimizer_generator.zero_grad()
if generator.label_type == 'none':
raise Exception('Incorrect Model: ACGAN generator must require labels')
if generator.label_type == 'required':
fake = generator(noise, labels)
elif generator.label_type == 'generated':
label_gen = torch.randint(0, generator.num_classes, (batch_size,), device=device)
fake = generator(noise, label_gen)
cgz = discriminator(fake, mode='classifier')
if generator.label_type == 'required':
loss = self.forward(cgz, labels)
else:
label_gen = label_gen.type(torch.LongTensor).to(device)
loss = self.forward(cgz, label_gen)
loss.backward()
optimizer_generator.step()
return loss.item()
[docs]class AuxiliaryClassifierDiscriminatorLoss(DiscriminatorLoss):
r"""Auxiliary Classifier GAN (ACGAN) loss based on a from
`"Conditional Image Synthesis With Auxiliary Classifier GANs
by Odena et. al. " <https://arxiv.org/abs/1610.09585>`_ paper
Args:
reduction (str, optional): Specifies the reduction to apply to the output.
If ``none`` no reduction will be applied. If ``mean`` the outputs are averaged over batch size.
If ``sum`` the elements of the output are summed.
override_train_ops (function, optional): A function is passed to this argument,
if the default ``train_ops`` is not to be used.
"""
def forward(self, logits, labels):
return F.cross_entropy(logits, labels, reduction=self.reduction)
[docs] def train_ops(self, generator, discriminator, optimizer_discriminator, real_inputs, device, labels=None):
r"""Defines the standard ``train_ops`` used by the Auxiliary Classifier discriminator loss.
The ``standard optimization algorithm`` for the ``discriminator`` defined in this train_ops
is as follows (label_g and label_d both could be either real labels or generated labels):
1. :math:`fake = generator(noise, label_g)`
2. :math:`value_1 = classifier(fake, label_g)`
3. :math:`value_2 = classifier(real, label_d)`
4. :math:`loss = loss\_function(value_1, label_g) + loss\_function(value_2, label_d)`
5. Backpropagate by computing :math:`\nabla loss`
6. Run a step of the optimizer for discriminator
Args:
generator (torchgan.models.Generator): The model to be optimized. For ACGAN, it must require labels
for training
discriminator (torchgan.models.Discriminator): The discriminator which judges the
performance of the generator.
optimizer_discriminator (torch.optim.Optimizer): Optimizer which updates the ``parameters``
of the ``discriminator``.
real_inputs (torch.Tensor): The real data to be fed to the ``discriminator``.
device (torch.device): Device on which the ``generator`` and ``discriminator`` is present.
batch_size (int): Batch Size of the data infered from the ``DataLoader`` by the ``Trainer``.
labels (torch.Tensor, optional): Labels for the data.
Returns:
Scalar value of the loss.
"""
if self.override_train_ops is not None:
return self.override_train_ops(generator, discriminator, optimizer_discriminator,
real_inputs, device, labels)
if labels is None:
raise Exception('ACGAN Discriminator requires labels for training')
if generator.label_type is 'none':
raise Exception('Incorrect Model: ACGAN generator must require labels for training')
batch_size = real_inputs.size(0)
noise = torch.randn(batch_size, generator.encoding_dims, device=device)
optimizer_discriminator.zero_grad()
cx = discriminator(real_inputs, mode='classifier')
if generator.label_type == 'required':
fake = generator(noise, labels)
elif generator.label_type == 'generated':
label_gen = torch.randint(0, generator.num_classes, (batch_size,), device=device)
fake = generator(noise, label_gen)
cgz = discriminator(fake, mode='classifier')
if generator.label_type == 'required':
loss = self.forward(cgz, labels) + self.forward(cx, labels)
else:
label_gen = label_gen.type(torch.LongTensor).to(device)
loss = self.forward(cgz, label_gen) + self.forward(cx, labels)
loss.backward()
optimizer_discriminator.step()
return loss.item()