StyleGAN 2 Truncation
StyleGAN 2 trained on images of landscapes, with varying levels of truncation.
At the beginning, all images have been fully truncated, showing the "average" landscape of all generated landscapes. When ψ = 1, no truncation is applied, showing full diversity across the distribution of images. When ψ = 1.5, generated images can reach beyond its own distribution, lowering image quality significantly, but increasing diversity beyond what is even in the real distribution.
Code: https://github.com/manicman1999/Style...
Research Paper: Analyzing and Improving the Image Quality of StyleGAN
StyleGAN, Tensorflow 2.0, Truncation, Landscapes