Architecture
aka generative adversarial network
GAN
Two-network setup where a generator and a discriminator compete during training.
Definition
GANs train a generator to fool a discriminator that distinguishes real from synthetic data. The result is sharp samples that, before diffusion, set the standard for image generation. StyleGAN remains relevant for faces and high-fidelity domains.
Common use cases
- Face generation
- Image super-resolution
- Domain transfer