5 Selecting characteristics in generated images

This chapter covers

  • Building a conditional generative adversarial network to generate images with certain attributes (human faces with or without eyeglasses, for example)
  • Implementing Wasserstein distance and gradient penalty to improve image quality
  • Selecting vectors associated with different features so that the trained GAN model generates images with certain characteristics (male or female faces, for example)
  • Combining conditional GAN with vector selection to specify two attributes simultaneously (female faces without glasses or male faces with glasses, for example)

The anime faces we generated with deep convolutional GAN (DCGAN) in chapter 4 look realistic. However, you may have noticed that each generated image has different attributes such as hair color, eye color, and whether the head tilts toward the left or right. You may be wondering if there is a way to tweak the model so that the generated images have certain characteristics (such as with black hair and tilting toward the left). It turns out you can.

5.1 The eyeglasses dataset

5.1.1 Downloading the eyeglasses dataset

5.1.2 Visualizing images in the eyeglasses dataset

5.2 cGAN and Wasserstein distance

5.2.1 WGAN with gradient penalty

5.2.2 cGANs

5.3 Create a cGAN

5.3.1 A critic in cGAN

5.3.2 A generator in cGAN

5.3.3 Weight initialization and the gradient penalty function

5.4 Training the cGAN

5.4.1 Adding labels to inputs

5.4.2 Training the cGAN

5.5 Selecting characteristics in generated images

5.5.1 Selecting images with or without eyeglasses

5.5.2 Vector arithmetic in latent space

5.5.3 Selecting two characteristics simultaneously

Summary