Generating Realistic Images Using Generative Adversarial Networks

3 min readNov 10, 2020

A generative adversarial network (GAN) is a class of machine learning frameworks designed by Ian Goodfellow and his colleagues in 2014. Two neural networks contesting with each other in a game (in the sense of game theory, often but not always in the form of a zero-sum game). Given a training set, this technique learns to generate new data with the same statistics as the training set.

GANs consists of two networks that compete with each other namely the generator network and discriminator network, discriminator network is designed in such a way that it can distinguish between real and fake data whereas the generator network is designed in such a way that it can produce fake data so that it can fool discriminator network.

Generally, a latent vector (random noise) is given as input to the generator network to generate fake images and these images are mixed with real images and given as input to the discriminator network to train it to distinguish between real and fake data, based on the output of discriminator our generator network learns accordingly how to make fake data that are close enough to fool discriminator and this is a never-ending process and also we cannot guarantee that after each step generator gets better always i.e. their loss functions keeps on fluctuating.

Generative adversarial networks are still developing and are getting better and better every year starting from deep convolutional GANs to StyleGAN we can see enormous changes in their outputs as well as their neural networks.

USE CASES OF GENERATING REALISTIC IMAGES:

✇ To generate fashion images useful for a designer to design clothes, shoes, jewelry, etc with ease.