We present a novel approach for the interactive creation of procedural 3D models in a more efficient way than state of the art tools for 3D reconstruction. Our tool uses the split grammar approach in an interactive setting and generalizes it in several ways. The tool allows for code-free, direct manipulation in 3D, and we provide operations to merge and specialize subdivided scopes. This remedies a severe limitation of context-free grammars, which can only produce a tree of splits. The user can interactively browse through the refinement hierarchy, apply rules and change parameters visually. We demonstrate the effectiveness of our tool with a reconstruction of some façade parts of a prime example of classical architecture, the Louvre in Paris.

Complex man-made shapes exhibit great regularities for a number of reasons, from functionality over manufacturability to aesthetics and style. A procedural representation is therefore commonly perceived as most appropriate. The problem, however, is to create it in an efficient, user-friendly way. One alternative is to use a programming language like Python (Blender), JavaScript (e.g., with WebGL), or CGAShape (grammar language of the CityEngine). Programmed models can be quite scalable – with CGA Shape for instance, whole cities can be described – but not so many 3D artists or CH professionals accept a code editor as user interface for 3D modeling; and only few of them are good programmers.

In order to save the virtues and remedy the deficiencies of procedural modeling we have developed a new modeling tool. The Louvre was selected as target model because it poses serious challenges:

• Classical architecture is very procedural, with many levels of nested repetitions and symmetries.
• Domain coverage: As a masterpiece, the Louvre covers most of the design space of the classical architecture that was highly influential all over Europe
• Scalability: The Louvre is huge (600m long, 300 mwide) and highly detailed. Purely manual reconstruction will hardly succeed even with excessive copy & paste.
• Irregularities: Seemingly regular parts are irregular because of 800 years of construction history. A post facto correction of inaccurate model parameters is indispensable to avoid costly re-modeling.

On a more abstract level, the challenge is to find a bridge between the flexibility of programming and the intuitiveness of direct 3D interaction. We propose a modeling by example method, providing a small but well-defined set of modeling operations. The research question is: Are they sufficient for the domain of classical architecture?
The benefit for the CH community is a more efficient 3D modeling toolkit than classical forward modelers (SketchUp etc.) without the overhead of coding (CityEngine). The tool will undergo first user tests during summer/autumn 2011.