Material Design via Neuroevolution

home | applications | research | people

abstract

The artistic process of developing models with complex appearances requires the support of intuitive, interactive tools. While spatially varying bidirectional reflection distribution functions (SBRDFs) provide control over appearances that vary across a model's surface, current techniques for their creation are largely limited to measuring and representing real-world materials. In contrast, we present an intuitive, brush-based visual interface that allows a user to design materials with only sparse input by painting desired light and dark regions onto objects from arbitrary viewpoints. This input guides the evolution of an optimized SBRDF that is consistent with these constraints, while maintaining physical plausibility for unspecified points and viewing directions. The keys to interactivity are a novel GPU-based real-time importance sampling algorithm and a fast neuroevolution-driven optimization scheme explicitly designed for this problem domain.

people

publications

video

  • Video (DiVX/31 MB) - Walk through demonstration of SHINE (Surface Highlighting via Interactive NeuroEvolution)