Computer Graphics World

December 2008 22 ■ ■ ■ ■ Virtual Architecture Frischer recalls. "Arches that should have been rounded were oval-shaped or worse." At the time, there was no algorithmic method of correcting those distortions using standard post-processing tools. So, two of Guidi's grad students developed a solution that would crawl through the scan data looking for repeated geometric forms based on volume and form. "I assumed that Gismondi's team had created the physi- cal models using repeated elements," says Frischer. He was correct: 99 percent of the buildings correspond to 38 simple forms, and when mixed and matched, created the impression of extensive variety. Using Maya and Creator, the group crafted new handmade models from digital versions of the forms, and then several of Guidi's students began the process of swap- ping out the messy scan data for the clean models. Textures were painstakingly applied by hand. Interactivity was added to the over- all model via MultiGen's OpenFlight. In June 2007, Rome Reborn 1.0 de- buted. It featured two classifi cations of buildings: Class 1, which are supported by hard data and reconstructed with 100 percent accuracy; and Class 2, buildings and infrastructure, like bridges, canals, and roads, that archaeologists know existed but can only approximate their location and aesthetic. Nearly 300 components of the city were Class 1 models, but Frischer's team had the time and funding to re-create only 31 of them with full detail. So many Class 1s, along with most of the Class 2s, had textures but lacked architectonic detail. Some surfaces didn't even have textures. " e results were acceptable when seen from afar, but do not support a close view, as happens in a virtual-reality session when you want to simulate the experience of walking down a Roman street," says Frischer. At a total of nine million triangles, the ver- sion contained a number of fi ll-in models, basic geometric building forms with some texture detail; these were intended to supple- ment the main pieces, including the Forum and Colosseum. Painted vertexes atop virtual terrain maps gave the cityscape life. Furthermore, Rome Reborn 1.0 was mobile; it ran on a suped-up Shuttle work- station with an Nvidia 8800 card, enabling the user to move around the model at 20 frames per second. Impressive? Yes. But aside from the basic fi ll-in models, there was another component missing—it could not be published via the Internet as a real- time interactive model. Rome, In All Its Glory Not resting on its laurels, the group contin- ued to pursue Frischer's ultimate goals for the project. e crew's fi rst objective with Rome Reborn 2 was to upgrade the detail and resolution of the structures. Frischer had recalled hearing about a unique urban-mod- eling solution from Pascal Mueller, who, as a graduate student at ETH Zurich, presented a SIGGRAPH paper in 2006 detailing how he used his technology—the fi rst known script- ing language specialized for producing archi- tectural content—to reconstruct Pompeii. e timing was perfect. It was the spring of 2008, and Mueller was setting up his company, Procedural, to sell the City Engine software (which was offi cially released this past August at SIGGRAPH 2008). CityEngine enables users to model entire cities as much as 10 times faster than with previous solutions. It contains a number of procedural modeling tools for generating large-scale urban layouts. At the software's core is unique shape grammar for effi - ciently creating detailed building models; the grammar supports a rules-based pro- duction scheme of architectural content at arbitrary levels of detail. As a result, it can produce as many buildings and objects as desired, in any style or mix of styles. Rather than creating a building by hand, which can take upward of a month, the software automatically reconstructs a mod- el by randomizing the windows, facades, doorways, and so forth (in geometric form) based on preset directives—akin to what Guidi's students did in Rome Reborn 1.0. Moreover, the user can choose the resolu- tion of the fi nal output, which can change depending on the application—whether it's low resolution for the Web or high reso- lution for an off -line rendering. And, as new archaeological fi ndings occur, the en- tire model can be updated easily to refl ect the discoveries. "We never had that kind of fl exibility before with Maya or 3ds Max models," Frischer points out. While it took a few months to set up the building rules in CityEngine, the Proce- dural group spent less than two weeks on the actual content—something Mueller es- timates would have taken two man-years to complete without CityEngine. Each model is generated from geometry in the asset li- brary, so there are no baked-in textures for the higher-resolution imagery, which aver- Using Procedural's CityEngine software, the group is upgrading the general structures with high- res models in record time, freeing others to focus their attention on the unique monuments. Images courtesy Procedural.

• Cover