Visual parameter exploration in GPU shader space

Abstract

The wide availability of high-performance GPUs has made the use of shader programs in visualization ubiquitous. Understanding shaders is a challenging task. Frequently it is difficult to mentally reconstruct the nature and types of transformations applied to the underlying data during the visualization process. We propose a method for the visual analysis of GPU shaders, which allows the flexible exploration and investigation of algorithms, parameters, and their effects. We introduce a method for extracting feature vectors composed of several attributes of the shader, as well as a direct manipulation interface for assigning semantics to them. The user interactively classifies pixels of images which are rendered with the investigated shader. The two resulting classes, a positive class and a negative one, are employed to steer the visualization. Based on this information, we can extract a wide variety of additional attributes and visualize their relation to this classification. Our system allows an interactive exploration of shader space and we demonstrate its utility for several different applications.