News
Youtube Channel CG
Dr.Ing. Iurie Chiosa
Universität Siegen
Lehrstuhl für Computergrafik
57076 Siegen
Current Research Areas:
Mesh Coarsening

Modern 3D model acquisition systems, e.g. 3Dscanners, or modeling systems can nowadays provide models with up to several millions of triangles. For a wide range of applications these models are too complex (in terms of the number of vertices) to be used. Thus, given the final budget of vertices, the problem is to simplify the input mesh by maintaining its original fidelity.
This work presents a mesh coarsening algorithm which is intended to capture the mesh features as good as possible. Three different types of the cluster's weights are proposed and the notion of Multiplicatively Weighted Centroidal Voronoi Diagram is used. To guarantee cluster connectivity, a new vertexboundarycount approach is introduced. Along with that, a new kneighborhood initialization algorithm for initial seeds generation is proposed.
Multilevel Mesh Clustering
demo video 
Two major classes of methods which are frequently employed for mesh clustering (segmentation, partitioning) are: Variational (Iterative) and Hierarchical. The result and the convergence of variational algorithm is strictly related to the initial number and selection of the seeds. On the other hand, hierarchical approaches have a greedy nature, i.e. a nonoptimal shape of the clusters in the hierarchy.
In this work a novel and generic Multilevel (ML) Mesh Clustering algorithm is proposed. The algorithm incorporates the advantages and resolves the inherent problems of both hierarchical and variational (Lloyd) algorithms. As a result, the initial number of seeds is not predefined and on each level the obtained clustering configuration is quasioptimal. The algorithm is able to perform a complete mesh analysis regarding the underlying energy functional. To reconstruct any level in the ML construction, a special incremental data structure is also proposed in this context.
GPUbased Mesh Clustering
demo video 
demo video 
demo video 
The use of parallel hardware, i.e. the graphics processing unit (GPU), is a reasonable and crucial task for fast and qualitative clustering of large polygonal surfaces. However, there is little work done in this direction mainly due to the fact that most of the algorithms have a sequential nature, they require a Half Edge data structure and use a global Priority Queue.
In this work two new parallel mesh clustering concepts are proposed together with an efficient GPUbased implementation. To perform the clustering on the GPU a specific mesh representation is also proposed. Using this framework we show that considerable speedup can be obtained.