ஜர்னல் ஆஃப் மாஸ் கம்யூனிகேஷன் & ஜர்னலிசம்

Combining Texton Broadcasting with Noise Injection in Stylegan-2 to Create a More Universal Method of Texture Synthesis

Abstract

Philip Riepe*

In the StyleGAN-2 framework, a novel multiscale texton broadcasting module is incorporated into our universal texture synthesis strategy. The texton broadcasting module adds an inductive bias, allowing for the creation of more textures, from those with regular structures to those that are completely random. We create a comprehensive, high-resolution dataset called NUUR-Texture500 to train and evaluate the proposed method. This dataset includes both the variety of natural textures and the stochastic variations that occur within each perceptually uniform texture. The results of the experiments show that the proposed method produces textures of significantly higher quality than the current state of the art. The complete comprehension of texture space is the ultimate objective of this work.