Category: GAN

#009 Developing a DCGAN for MNIST Dataset

Highlights: In the previous posts we have already explored the basic GAN idea, and we have studied the guidelines for more stable training. In particular, we have analyzed the “GAN Hacks” in post 007 that was proposed in a DCGAN paper. In addition, we have implemented a simple GAN network to learn the mapping of a 1D function.  Hence, to prove that the GANs are a promising family of generator architecture, we need to start…
Read more

#014 Pix2Pix Generative Adversarial Networks

Highlights: In the previous post, we introduced a way of training GAN models using a labeled dataset, known as Conditional Adversarial Networks (CGANs). CGANs are a revolution in the field of AI and are gaining popularity quite rapidly. One of the most famous GANs being used today is the Image2Image GAN or the Pix2Pix GAN. In this post, we will study Pix2Pix GAN in detail. Let’s begin. Tutorial overview: What is a Pix2Pix GAN? The…
Read more

#013 Conditional Generative Adversarial Networks (CGANs)

Highlights: In the previous posts, we studied Deep Convolutional Generative Adversarial Networks (DCGAN). These types of conventional DCGAN architectures are trained in an unsupervised and unconditional manner. Therefore, there are no labels involved in the training process. In this post, we will see how we can train our GAN model using a labeled dataset by the use of Conditional Generative Adversarial Networks (CGAN). So, let’s begin. Tutorial overview: What are Conditional GANs? Initializing and Defining…
Read more

#012 Understanding Latent Space in Generators

Highlights: Hello and welcome. In the previous posts, we have seen the architecture, building, and implementation of three DCGAN models using different datasets. By now, you must have gotten a fair bit of idea about Generative Adversarial Networks in general. In this post, we will go deeper into the workings of a generator in a typical GAN model. More specifically, we will understand the Latent Space which is used as an input for the generator.…
Read more

#011 Developing a DCGAN for CelebA Dataset

Highlights: In the previous two chapters, we built two distinct DCGAN models, one for MNIST Handwritten Digit dataset, and the other for CIFAR-10 dataset. In this chapter, we will build yet another DCGAN model. However, this time, we will use a different dataset known as the CelebA dataset. Let’s start by learning a bit about the CelebA dataset and its features. Tutorial overview: Downloading CelebA Dataset Initializing and Defining the DCGAN Model Training the DCGAN…
Read more