Perfect bounding for optimized evaluation of procedurally-generated scene data

A method of rendering a computer generated image including procedurally generated geometry, the method comprising:

• receiving scene data including a procedure adapted to generate scene geometry upon execution;
• in response to determining that a first bounding box encompassing a maximum potential size and associated with the procedure is potentially visible in an image to be rendered from the scene data:
• executing the procedure to generate a first instance of the scene geometry;
• partitioning the first instance of the scene geometry into at least first and second subsets of scene geometry;
• defining a second bounding box encompassing the first subset of the scene geometry;
• defining a third bounding box encompassing the second subset of the scene geometry;
• in response to determining that the second bounding box is potentially visible in the image, including the first subset of the scene geometry in scene data;
• in response to determining that the third bounding box is potentially visible in the image, including the second subset of the scene geometry in scene data;
• and
• rendering the image from the scene data.