WISE: Whitebox Image Stylization by Example-Based Learning

AbstractImage-based artistic rendering can synthesize a variety of expressive styles using algorithmic image filtering. In contrast to deep learning-based methods, these heuristics-based filtering techniques operate on high-resolution images, are interpretable, and be parameterized according various design aspects. However, adapting or extending produce new is often tedious error-prone task that requires expert knowledge. We propose paradigm alleviate this problem: implementing as differentiable operations learn parametrizations aligned certain reference styles. To end, we present WISE, an example-based image-processing system handle multitude stylization techniques, such watercolor, oil cartoon stylization, within common framework. By training parameter prediction networks for global local filter parameterizations, simultaneously adapt effects content, e.g., enhance facial features. Our method optimized in style-transfer framework learned generative-adversarial setting image-to-image translation. demonstrate jointly XDoG CNN postprocessing achieve comparable results state-of-the-art GAN-based method. https://github.com/winfried-loetzsch/wise.