Grabner, HaraldUllrich, TorstenFellner, Dieter W.Reinhard Klein and Pedro Santos2014-12-162014-12-162014978-3-905674-75-0https://doi.org/10.2312/gch.20141317In this paper we present a novel 3D model retrieval approach based on generative modeling techniques. In our approach generative models are created by domain experts in order to describe 3D model classes. These generative models span a shape space, of which a number of training samples is taken at random. The samples are used to train content-based retrieval methods. With a trained classifier, techniques based on semantic enrichment can be used to index a repository. Furthermore, as our method uses solely generative 3D models in the training phase, it eliminates the cold start problem. We demonstrate the effectiveness of our method by testing it against the Princeton shape benchmark.H.3.3 [Computer Graphics]Information SystemsInformation Search and RetrievalI.2.4 [Computer Graphics]Knowledge Representation Formalisms and MethodsRepresentations (procedural and rulebased)I.4.8 [Computer Graphics]Scene AnalysisObject recognition