StyleGAN is a GAN design released by researchers at NVIDIA in December 2018. It is essentially an upgraded version of ProGAN. It combined ProGAN with neural style transfer. At the core of the StyleGAN architecture is a style-transfer technique. The model set a new record for face generation tasks and can also be used to generate realistic images of cars, bedrooms, houses, and so on.
As with ProGAN, StyleGAN generates images gradually by starting with a very low resolution and continuing to a high-resolution image. The GAN controls the visual features that are expressed in each level, from coarse features such as the pose and face shape, through to the finer features such as eye and hair color:
The generator in StyleGAN incorporates a mapping network. The goal of the mapping network is to encode...
