Perfect Bounding for Optimized Evaluation of Procedurally-Generated Scene Data

Optimally-sized bounding boxes for scene data including procedurally generated geometry are determined by first determining whether an estimated bounding box including the procedural geometry is potentially visible in an image to be rendered. If so, the procedural geometry is generated and an optimal bounding box closely bounding the procedural geometry is determined. The generated procedural geometry may then be discarded or stored for later use. Rendering determines if the optimal bounding box is potentially visible. If so, then the associated procedural geometry is regenerated and rendered. Alternatively, after the estimated bounding box is potentially visible, the generated procedural geometry may be partitioned into subsets using a spatial partitioning scheme. A separate optimal bounding box is determined for each of the subsets. During rendering, if any of the optimal bounding boxes are potentially visible, the associated procedural geometry is regenerated and rendered.