Apply to speak at TensorFlow World. Deadline April 23rd. Propose talk

tf.contrib.gan.losses.combine_adversarial_loss

tf.contrib.gan.losses.combine_adversarial_loss(
    gan_loss,
    gan_model,
    non_adversarial_loss,
    weight_factor=None,
    gradient_ratio=None,
    gradient_ratio_epsilon=1e-06,
    scalar_summaries=True,
    gradient_summaries=True
)

Defined in tensorflow/contrib/gan/python/losses/python/tuple_losses_impl.py.

Combine adversarial loss and main loss.

Uses combine_adversarial_loss to combine the losses, and returns a modified GANLoss namedtuple.

Args:

  • gan_loss: A GANLoss namedtuple. Assume the GANLoss.generator_loss is the adversarial loss.
  • gan_model: A GANModel namedtuple. Used to access the generator's variables.
  • non_adversarial_loss: Same as main_loss from combine_adversarial_loss.
  • weight_factor: Same as weight_factor from combine_adversarial_loss.
  • gradient_ratio: Same as gradient_ratio from combine_adversarial_loss.
  • gradient_ratio_epsilon: Same as gradient_ratio_epsilon from combine_adversarial_loss.
  • scalar_summaries: Same as scalar_summaries from combine_adversarial_loss.
  • gradient_summaries: Same as gradient_summaries from combine_adversarial_loss.

Returns:

A modified GANLoss namedtuple, with non_adversarial_loss included appropriately.