4 Image generation with generative adversarial networks

This chapter covers

  • Designing a generator by mirroring steps in the discriminator network
  • How a 2D convolutional operation works on an image
  • How a 2D transposed convolutional operation inserts gaps between the output values and generates feature maps of a higher resolution
  • Building and training generative adversarial networks to generate grayscale and color images

You have successfully generated an exponential growth curve and a sequence of integers that are all multiples of 5 in chapter 3. Now that you understand how generative adversarial networks (GANs) work, you are ready to apply the same skills to generate many other forms of content, such as high-resolution color images and realistic-sounding music. However, this may be easier said than done (you know what they say: the devil is in the details). For example, exactly how can we make the generator conjure up realistic images out of thin air? That’s the question we’re going to tackle in this chapter.

4.1 GANs to generate grayscale images of clothing items

4.1.1 Training samples and the discriminator

4.1.2 A generator to create grayscale images

4.1.3 Training GANs to generate images of clothing items

4.2 Convolutional layers

4.2.1 How do convolutional operations work?

4.2.2 How do stride and padding affect convolutional operations?

4.3 Transposed convolution and batch normalization

4.3.1 How do transposed convolutional layers work?

4.3.2 Batch normalization

4.4 Color images of anime faces

4.4.1 Downloading anime faces

4.4.2 Channels-first color images in PyTorch

4.5 Deep convolutional GAN

4.5.1 Building a DCGAN

4.5.2 Training and using DCGAN

Summary