Join us at TensorFlow World, Oct 28-31. Use code TF20 for 20% off select passes. Register now

Module: tf.contrib.gan.eval

View source on GitHub

TF-GAN evaluation module.

This module supports techniques such as Inception Score, Frechet Inception distance, and Sliced Wasserstein distance.

Modules

classifier_metrics module: Model evaluation tools for TF-GAN.

eval_utils module: Utility file for visualizing generated images.

summaries module: Common TF-GAN summaries.

Functions

add_cyclegan_image_summaries(...): Adds image summaries for CycleGAN.

add_gan_model_image_summaries(...): Adds image summaries for real and fake images.

add_gan_model_summaries(...): Adds typical GANModel summaries.

add_image_comparison_summaries(...): Adds image summaries to compare triplets of images.

add_regularization_loss_summaries(...): Adds summaries for a regularization losses..

add_stargan_image_summaries(...): Adds image summaries to see StarGAN image results.

classifier_score(...): Classifier score for evaluating a conditional generative model.

classifier_score_from_logits(...): Classifier score for evaluating a generative model from logits.

diagonal_only_frechet_classifier_distance_from_activations(...): Classifier distance for evaluating a generative model.

frechet_classifier_distance(...): Classifier distance for evaluating a generative model.

frechet_classifier_distance_from_activations(...): Classifier distance for evaluating a generative model.

get_graph_def_from_disk(...): Get a GraphDef proto from a disk location.

get_graph_def_from_resource(...): Get a GraphDef proto from within a .par file.

get_graph_def_from_url_tarball(...): Get a GraphDef proto from a tarball on the web.

image_grid(...): Arrange a minibatch of images into a grid to form a single image.

image_reshaper(...): A reshaped summary image.

kernel_classifier_distance(...): Kernel "classifier" distance for evaluating a generative model.

kernel_classifier_distance_and_std(...): Kernel "classifier" distance for evaluating a generative model.

kernel_classifier_distance_and_std_from_activations(...): Kernel "classifier" distance for evaluating a generative model.

kernel_classifier_distance_from_activations(...): Kernel "classifier" distance for evaluating a generative model.

mean_only_frechet_classifier_distance_from_activations(...): Classifier distance for evaluating a generative model from activations.

preprocess_image(...): Prepare a batch of images for evaluation.

run_image_classifier(...): Runs a network from a frozen graph.

run_inception(...): Run images through a pretrained Inception classifier.

sliced_wasserstein_distance(...): Compute the Wasserstein distance between two distributions of images.

Other Members

  • INCEPTION_DEFAULT_IMAGE_SIZE = 299
  • frechet_inception_distance
  • inception_score
  • kernel_inception_distance
  • kernel_inception_distance_and_std