Chapter 1: Deep Learning Perceptron Chapter Goal: In this chapter we introduce the basics of deep learning from the perceptron to multi-layer perceptron. No of pages: 30 Sub -Topics 1. Understanding deep learning and supervised learning. 1. Using the perceptron for supervised learning. 2. Constructing a multilayer perceptron. 3.
Discover the basics of activation, loss, optimization and back propagation for problems of regression and classification. Chapter 2: Unleashing Autoencoders and Generative Adversarial Networks Chapter Goal: This chapter introduces the autoencoder and GAN for simple content generation. Along the way we also learn about using convolutional network layers for better feature extraction. No of pages: 30 Sub - Topics 1. Why we need autoencoders and how they function. 2. Improving on the autoencoder with convolutional network layers. 3.
Generating content with the GAN. 4. Explore methods for improving on the vanilla GAN. Chapter 3: Exploring the Latent Space Chapter Goal: In this chapter we discover the latent space in AI. What it means to move through the AI latent space using variational autoencoders and conditional GANs. No of pages : 30 Sub - Topics: 1. Understanding variation and the variational autoencoder. 2.
Exploring the latent space with a VAE. 3. Extending a GAN to be conditional. 4. Generate interesting foods using a conditional GAN. Chapter 4: GANs, GANs and More GANs Chapter Goal: In this chapter we begin uncovering the vast variations in GANs and their applications. We start with basics like the double convolution GAN and work up to the Stack and Progressive GANs. No of pages: 30 Sub - Topics: 1.
Look at samples from the many variations of GANs. 2. Setup and use a DCGAN. 3. Understand how a StackGAN works. 4. Work with and use a ProGAN. Chapter 5: Image to Image Translation with GANs Covers: Pix2Pix and DualGAN, side projects for understanding with ResNET and UNET, advanced network architectures for image classification/generation Chapter 6: Translating Images with Cycle Consistency Covers: Cycle consistency loss and the CycleGAN, BiCycleGAN and StarGAN Chapter 7: Styling with GANs Covers: StyleGAN, Attention and the Self-attention GAN with a look at DeOldify Chapter 8: Developing DeepFakes Chapter Goal: DeepFakes are taking the world by storm and in this chapter, we explore how to use a DeepFakes project.
No of pages: 30 1. Learn how to isolate faces or other points of interest in images or video. 2. Extract and replace faces from images or video. 3. Use DeepFakes GAN to generate facial images based on input image. 4. Put it all together and allow the user to generate their own DeepFake video.
Chapter 9: Uncovering Adversarial Latent Autoencoders Chapter Goal: GANs are not the only technique that allows for content manipulation and generations. In this chapter we look at the ALAE method for generating content. No of pages: 1. Look at how to extend autoencoders for adversarial learning. 2. Understanding how AE can be used to explore the latent space in data. 3. Use ALAE to generate conditional content.
4. Revisit our previous foods example and see what new foods we can generate. Chapter 10: Video Content with First Order Model Motion Chapter Goal: In this chapter we explore a new technique for animating static images called First Order Model Motion. At the end of this chapter we will use this technique to create avatars for Skype or Zoom. No of pages: 30 1. Discover the basic of First Order Model Motion, what it is and how it works. 2. Be able to apply FOMM to a number of static image datasets for various applications.
3. Use the project Avatarify for generating real-time avatars from static avatars. 4. Use Avatarify real-time in applications like Zoom or Skype.