View-Region Optimized Image-Based Scene Simplification

We present an algorithm for image based simplification of computer generated scenes for virtual reality (VR) applications. Our system transforms geometrically-complex environments into a layered quad tile representation renderable on low power mobile-class VR devices. A novel constrained optimization formulation ensures that the resulting tiles are renderable within a predetermined compute budget, limiting both primitive-count and fill rate. Furthermore, a new method for texturing from RGBD images generates high-quality silhouettes without the drawbacks of traditional point splatting. We demonstrate the effectiveness of this layered tile representation at rendering complex scenes with visually compelling, anti-aliased results.

• Machine Perception