Taejung Park

CUDA-based Signed Distance Field Calculation for Adaptive Grids

We present a simple yet robust signed distance field (SDF) generator based on recent GPU architectures. In our approach, the squared Euclidean distance is calculated for each triangle face in parallel, and then an optimized stream reduction process is used to find the shortest distance. During this process, the stream reduction operation acts like a parallel binary space-searching routine for each level. This process uses computations and memory bandwidth efficiently because of the massive number of CUDA threads. Signs are then determined by calculating angleweighted pseudonormals. Unlike some previous SDF approaches that only calculate the SDF near the surface or within the bounding box, our method can calculate the SDF adaptively so that there are no limitations on proximity or regularity. We also compare our GPU-based results with a kd-tree based single CPU approach for a 3D geometry synthesis application.

Automatic player behavior analysis system using trajectory data in a massive multiplayer online game

This paper presents a new automated behavior analysis system using a trajectory clustering method for massive multiplayer online games (MMOGs). The description of a player’s behavior is useful information in MMOG development, but the monitoring and evaluation cost of player behavior is expensive. In this paper, we suggest an automated behavior analysis system using simple trajectory data with few monitoring and evaluation costs. We used hierarchical classification first, then applied an extended density based clustering algorithm for behavior analysis. We show the usefulness of our system using trajectory data from the commercial MMOG World of Warcraft (WOW). The results show that the proposed system can analyze player behavior and automatically generate insights on players’ experience from simple trajectory data.

RViz: a toolkit for real domain data visualization

In computational science and computer graphics, there is a strong requirement to represent and visualize information in the real domain, and many visualization data structures and algorithms have been proposed to achieve this aim. Unfortunately, the dataflow model that is often selected to address this issue in visualization systems is not flexible enough to visualize newly invented data structures and algorithms because this scheme can accept only specific data structures. To address this problem, we propose a new visualization tool, RViz, which is independent of the input information data structures. Since there is no requirement for additional efforts to manage the flow networks and the interface to abstracted information is simple in RViz, any scientific information visualization algorithms are easier to implement than the dataflow model. In this paper, we provide case studies in which we have successfully implemented new data structures and related algorithms using RViz, including geometry synthesis, distance field representation, and implicit surface reconstruction. Through these cases, we show how RViz helps users visualize and understand any hidden insights in input information.

Primitive Trees for Precomputed Distance Queries

We propose the primitive tree, a novel and compact space‐partition method that samples and reconstructs a distance field with high accuracy, even for regions far from the surfaces. The primitive tree is based on the octree and stores the indices of the nearest primitives in its leaf nodes. Most previous approaches have involved a trade‐off between accuracy and speed in distance queries, but our method can improve both aspects simultaneously. In addition, our method can sample unsigned distance fields effectively, even for self‐intersecting and nonmanifold models. We present test results showing that our method can sample and represent large scenes, with more than ten million triangles, rapidly and accurately.

Analytic Solutions of Integral Moving Least Squares for Polygon Soups

This paper presents analytic solutions to the integral moving least squares (MLS) equations originally proposed by Shen et al. by choosing another specific weighting function that renders the numerator in the MLS equation unitless. In addition, we analyze the original method to show that their approximation surfaces (i.e., enveloping surfaces with nonzero \epsilon values in the weighting function) often form zero isosurfaces near concavities behind the triangle-soup models. This paper also presents error terms for the integral MLS formulations against signed distance fields. Based on our analytic solutions, we show that our method provides both interpolation and approximation surfaces faster and more efficiently. Because our method computes solutions for integral MLS equations directly, it does not rely on numerical steps that might have numerical-accuracy issues. In particular, unlike the original method that deals with incorrect approximation surfaces by iteratively adjusting parameters, this paper proposes faster and more efficient approximations to surfaces without needing iterative routines. We also present computational efficiency comparisons, in which our method is 15-fold faster in computing integrations, even with conservative assumptions. Finally, we show that the surface normal vectors on the implicit surfaces formed by our analytic solutions are identical to the angle-weighted pseudonormal vectors.

Adaptive Synthesis of Distance Fields

We address the computational resource requirements of 3D example-based synthesis with an adaptive synthesis technique that uses a tree-based synthesis map. A signed-distance field (SDF) is determined for the 3D exemplars, and then new models can be synthesized as SDFs by neighborhood matching. Unlike voxel synthesis approach, our input is posed in the real domain to preserve maximum detail. In comparison to straightforward extensions to the existing volume texture synthesis approach, we made several improvements in terms of memory requirements, computation times, and synthesis quality. The inherent parallelism in this method makes it suitable for a multicore CPU. Results show that computation times and memory requirements are very much reduced, and large synthesized scenes exhibit fine details which mimic the exemplars.

Progressive Compression of Geometry Information with Smooth Intermediate Meshes

We present a new geometry compression algorithm for manifold 3D meshes based on octree coding. For a given mesh, regular volume grids are built with an adaptive octree. For each grid point, a binary sign, which indicates inside or outside of the mesh, is generated based on the distance to the mesh. In each leaf cell having a vertex, a least square fitting plane is created for a localized geometry range with signs. Finally, quantized geometry information is locally encoded. We demonstrate that the octree with signs can be used to predict the vertex positions. As a result, the proposed method generates competitive bitrates compared to the current state-of-art progressive geometry coder. Our method also shows better rate-distortion performance during decompression or transmission with improved smoothness.

Face mesh compression based on half-spherical coordinates for home entertainment systems

For a realistic entertainment experience, home entertainment systems require the ability to handle rich and detailed human facial expressions; this consumes huge amounts of storage. We present a progressive octree-based mesh compression framework optimized for human face models in game applications. We define the octree data structure in spherical coordinates (S-Octree) to reduce information entropy for human face models and to utilize the symmetry. Because our approach provides a framework for general spatial partitioning methods, it can be applied to other compression schemes based on octrees for human facial meshes. In comparison with previous methods, our method reduces number of bits around by 50%.