Architects, urban planners, the film and gaming industry as well as map makers are all very much interested in detailed, semantic city models. While generating a synthetic city has become much easier thanks to procedural modelling, automatically modelling an existing city remains a challenge. With procedural modelling, a building is described as a procedure, wherein an initial shape is successively refined according to a formal grammar set of rules. In this work, we present a set of methods to automatically instantiate and generate such sets of rules, from pictures. First, we present the applications and challenges behind city modelling. Shape grammars have greatly contributed to making city modelling easier. From there, a large spectrum of applications became possible, such as generating realistic large-scale virtual cities or fine-grained urban planning simulations. We introduce the shape grammar framework and explain how it is used to represent buildings. Nevertheless, when it comes to modelling an existing city, much of the work remains a tedious, manual task. The first steps to an automatic pipeline include facade extraction and style detection. Next, we distinguish

In this paper, we address the following research problem: How can we generate a meaningful split grammar that explains a given fa-cade layout? To evaluate if a grammar is meaningful, we propose a cost function based on the description length and minimize this cost using an approximate dynamic programming framework. Our eval-uation indicates that our framework extracts meaningful split gram-mars that are competitive with those of expert users, while some users and all competing automatic solutions are less successful.

We propose a new approach for semantic segmentation of 3D city models. Starting from an SfM reconstruction of a street-side scene, we perform classification and facade splitting purely in 3D, obviating the need for slow image-based semantic segmentation methods. We show that a properly trained pure-3D approach produces high quality labelings, with significant speed benefits (20x faster) allow-ing us to analyze entire streets in a matter of minutes. Addi-tionally, if speed is not of the essence, the 3D labeling can be combined with the results of a state-of-the-art 2D clas-sifier, further boosting the performance. Further, we pro-pose a novel facade separation based on semantic nuances between facades. Finally, inspired by the use of architec-tural principles for 2D facade labeling, we propose new 3D-specific principles and an efficient optimization scheme based on an integer quadratic programming formulation. 1.

(Belgium) Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenigvuldigd en/of openbaar gemaakt worden door middel van druk, fotokopie, microfilm, elektronisch of op welke andere wijze ook zonder voorafgaande schriftelijke toestemming van de uitgever. All rights reserved. No part of the publication may be reproduced in any form by print, photoprint, microfilm, electronic or any other means without written permission from the publisher. ISBN XXX-XX-XXXX-XXX-X D/XXXX/XXXX/XX

Mueller et al. [Müller et al. 2007] assume that the facade can be split into a single irregular grid of rectangles, called tiles. In each tile, there is preferably a single element, e.g., window or door. This works well for a large set of facades, but it does not work as well for more complex facades or facades modeled in greater detail (as is the case in our test dataset). If a facade does not conform to the assumed facade model, facade elements have to be split. We do not consider this as valid solution in our paper. Here are some examples of layouts that cannot be represented well: • facades with more than one grid of elements where the grids are not aligned • facades with interleaved grids • signs that span multiple floors or columns of elements • facades with ornaments, e.g., pilasters, between windows • facades with columns that span multiple floors • facades with high doors that span multiple floors • ornaments that go across the whole facade • ornaments that have a higher repetition frequency than the base grid of tiles, e.g., triglyphs In Fig. 1 we show examples that fit well into the model of Mueller et al. In Fig. 2 we show examples that do not work well. 2 Generating a single deterministic grammar vs. combining multiple deterministic gram-mars There are two problems that can be distinguished in our context. The first problem is the following: Given a single layout as input, how can we automatically extract a deterministic split grammar to represent the input layout? The second problem is the following: Given a set of deterministic split grammars that each describe a facade layout, how can we combine the deterministic split gram-mars to generate a stochastic split grammar? The stochastic split