Fig. 1
From: Application of generative adversarial networks (GAN) for ophthalmology image domains: a survey
An illustration of a basic architecture of GAN (vanilla GAN) for retinal image synthesis. The generator transforms a noise vector \(z\) from the distribution \(p(z)\) into a synthesized retinal image \({x}_{g}\). The discriminator distinguishes the synthetic and real retinal images based on the distributions of \({x}_{g}\) and \({x}_{r}\), respectively. The generated image samples form a distribution \({p}_{g}(x)\), which is desired to be an approximation of \({p}_{r}(x)\) from real image sample, after successful training