Module google/compare_gan/model_8_lsun_bedroom_resnet19/1

ResNet19 trained on LSUN Bedroom (FID: 42.51).

Module URL: https://tfhub.dev/google/compare_gan/model_8_lsun_bedroom_resnet19/1

Open Colab notebok

Overview

ResNet19 generator and discriminator.

For the details of the setup, please refer to [1]. The code used to train these models is available on GitHub. View all available compare_gan modules in the Colab notebook.

Details

  • Dataset: LSUN Bedroom
  • Model: Non-saturating GAN
  • Architecture: ResNet19
  • Optimizer: Adam (lr=2.851e-04, beta1=0.102, beta2=0.998)
  • Discriminator iterations per generator iteration: 1
  • Discriminator normalizaton: Spectral normalization
  • Discriminator regularization: none

Scores

  • FID: 42.51
  • Inception: 3.58
  • MS-SSIM: N/A

Example use

# Declare the module
gan = hub.Module("https://tfhub.dev/google/compare_gan/model_8_lsun_bedroom_resnet19/1")

# Use the generator signature
z_values = tf.random_uniform(minval=-1, maxval=1, shape=[64, 128])
images = gan(z_values, signature="generator")

# Use the discriminator signature
logits = gan(images, signature="discriminator")

# Drive execution with tf.Session
session.run([images, logits])

References

[1] Karol Kurach, Mario Lucic, Xiaohua Zhai, Marcin Michalski, Sylvain Gelly. The GAN Landscape: Losses, Architectures, Regularization, and Normalization.