An empirical comparison of four initialization methods for the K-Means algorithm
Pattern Recognition Letters
Vicinity Shading for Enhanced Perception of Volumetric Data
Proceedings of the 14th IEEE Visualization 2003 (VIS'03)
Local Histograms for Design of Transfer Functions in Direct Volume Rendering
IEEE Transactions on Visualization and Computer Graphics
Ambient Occlusion and Edge Cueing for Enhancing Real Time Molecular Visualization
IEEE Transactions on Visualization and Computer Graphics
A performance study of general-purpose applications on graphics processors using CUDA
Journal of Parallel and Distributed Computing
Clustering billions of data points using GPUs
Proceedings of the combined workshops on UnConventional high performance computing workshop plus memory access workshop
Accelerating K-Means on the Graphics Processor via CUDA
INTENSIVE '09 Proceedings of the 2009 First International Conference on Intensive Applications and Services
K-Means on Commodity GPUs with CUDA
CSIE '09 Proceedings of the 2009 WRI World Congress on Computer Science and Information Engineering - Volume 03
PACIFICVIS '09 Proceedings of the 2009 IEEE Pacific Visualization Symposium
The Occlusion Spectrum for Volume Classification and Visualization
IEEE Transactions on Visualization and Computer Graphics
EUROVIS'05 Proceedings of the Seventh Joint Eurographics / IEEE VGTC conference on Visualization
Viewpoint selection for intervention planning
EUROVIS'07 Proceedings of the 9th Joint Eurographics / IEEE VGTC conference on Visualization
Hi-index | 0.00 |
In the past years many interactive volume rendering techniques have been proposed, which exploit the neighboring environment of a voxel during rendering. In general on-the-fly acquisition of this environment is infeasible due to the high amount of data to be taken into account. To bypass this problem we propose a GPU preprocessing pipeline which allows to acquire and compress the neighborhood information for each voxel. Therefore, we represent the environment around each voxel by generating a local histogram (LH) of the surrounding voxel densities. By performing a vector quantization (VQ), the high number of LHs is than reduced to a few hundred cluster centroids, which are accessed through an index volume. To accelerate the required computational expensive processing steps, we take advantage of the highly parallel nature of this task and realize it using CUDA. For the LH compression we use an optimized hybrid CPU/GPU implementation of the k-means VQ algorithm. While the assignment of each LH to its nearest centroid is done on the GPU using CUDA, centroid recalculation after each iteration is done on the CPU. Our results demonstrate the applicability of the precomputed data, while the performance is increased by a factor of about 10 compared to previous approaches.