A new paper by Tero Karras, Samuli Laine, Timo Aila from NVIDIA called: “A Style-Based Generator Architecture for GANs” proposes an alternative generator architecture for generative adversarial networks, using style transfer. You will be surprised about the advances in GANs field. Using the Style-Based Generator Architecture for Generative Adversarial Networks by Tero Karras (NVIDIA), Samuli Laine (NVIDIA), Timo Aila (NVIDIA) After receiving all the images, fake and real, the Discriminator returns probabilities, a number in the range of 0 and 1, 1 representing a prediction of authenticity and 0 representing fake.The new generated image is passed to the Discriminator alongside with some images taken from the actual Dataset.Feed the Generator with a Dataset ( Example: celebrity faces), so it can return new images.If we can describe the process in few steps, it will be something like this: The goal of the Discriminator is to identify images coming from the Generator as fake ones. So, the final goal of the Generator is to create almost perfect cat’s pictures for the Discriminator without being caught. The Generator responsibility is to create new synthetic images that pass to the Discriminator, expecting that it will be classified as authentic, even when it’s fake. The goal of the Discriminator is to recognize which pictures belongs to the Cats dataset. The Discriminator takes the decision to classify the image as a real or Fake.Īs an example, let’s imagine we are trying to replicate cat’s pictures, using the Cats dataset. The Generator creates new images while the Discriminator evaluate if they are real or fake. Like I said before, GAN’s architecture consists of two networks: Discriminator and Generator. The field of application is getting bigger, you can use GANs for the next purposes: That led me to read more about what GANs can do and the uses cases are really interesting.

Celebrity fake news generator