In procedural modeling, a single rule set can produce a wide variety of 3D models (teaser image left). This paper presents a thumbnail gallery generation system which automatically samples a rule set, clusters the resulting models into distinct groups (teaser image middle), and selects a representative image for each group to visualize the diversity of the rule set (teaser image right).
Computer Graphic Forum (Proceedings of Eurographics), 2014
Procedural modeling allows for the generation of innumerable variations of models from a parameterized, conditional or stochastic rule set. Due to the abstractness, complexity and stochastic nature of rule sets, it is often very difficult to have an understanding of the diversity of models that a given rule set defines. We address this problem by presenting a novel system to automatically generate, cluster, rank, and select a series of representative thumbnail images out of a rule set. We introduce a set of ‘view attributes’ that can be used to measure the suitability of an image to represent a model, and allow for comparison of different models derived from the same rule set. To find the best thumbnails, we exploit these view attributes on images of models obtained by stochastically sampling the parameter space of the rule set. The resulting thumbnail gallery gives a representative visual impression of the procedural modeling potential of the rule set. Performance is discussed by means of a number of distinct examples and compared to state-of-the-art approaches.
Secord et al.’s [SLF∗11] view attribute combination (left) versus ours (right). Circles denote differences between both methods, dashed circles stand for minor differences only. The bar charts show the user preference counts: left is ours, middle is Secord et al., right is no preference.
Thumbnail galleries for three rule sets. Left: Philadelphia, center: Paris-style building, right: procedural street.