ShapeGraMM: On the fly procedural generation of massive models for real-time visualization

Abstract

In this study, we propose ShapeGraMM (Shape Grammar for massive models), an expandable grammar that procedurally generates geometries in real-time to create 3D scenes of massive models. Procedural modeling has attracted attention for its ability to quickly create 3D scenes using a compact representation, which stores generation rules rather than an explicit representation of the scene. Our work is an extension of the Computer Generated Architecture (CGA) shape grammar specification. We introduce rules that will take the visibility of the camera into account and decide whether or not it should continue to generate. ShapeGraMM explores the repetitions and patterns present in massive models, helping it render scenes, reduce its memory footprint, and procedurally process the scene efficiently. We propose an engine implementation that generates scenes on the fly using ShapeGraMM, followed by its evaluation using massive real-world models. We conducted a case study testing our solution with large-scale CAD models of the oil & gas industry in the web context, which is a more limited platform than the desktop. In contrast to random models commonly generated by procedural generation methodologies, the original models of our study were modeled in a non-random method. Our results show that the solution has high performance and maintains a compact representation of the models offline and in runtime. The procedural engine renders a massive scene containing over 11 million objects at interactive frame rates.