INTRODUCTION

We address the automatic generation of large geometric models. This is a key issue in visualization for two reasons. First, it is a non-trivial task to create an interesting model of large size. Second, current challenges in visualization are mostly related to the largeness of the data sets. Tabular or raster data are usually generated by physical devices (cameras, satellites, etc) and for these, it is not hard to get large samples. But geometric data sets generally require human model effort, and can be a painstaking task without the proper tools. One source of architectural models might be those already designed with CAD tools and stored in DXF format. But even to construct a geometric model from such files can be a challenging task [vanWyk-Ritchie].

Our goal is the statistical generation of large geometric models which can be useful for visualization experimentation. In the following, our basic example is the problem of generating a city environment resembling Manhattan. Such a model might be useful for visualization fly-overs and down to walk throughs, or for animation. There are two issues:

1. We need ways of specifying large scale models without coding each detail by hand. E.g. what tools do we need to conveniently generate a city neigbhood comprising about 50 city blocks?
2. We must parameterize our model generation so that, by setting different options, we can generate different versions of a basic design. The choice and design of the parameter space is critical.

In this note, we hope to break down this problem to a number of relatively independent tasks, so that the entire class can participate in this effort to generate one large model. Note that our objective is to get this model, not its visualization (that is another issue we want to address). Still, it is unavoidable in the project to provide some simple tools to view or validate your model parts.