Fast uncertainty-driven large-scale volume feature extraction on desktop PCs

The ability to efficiently and accurately extract features of interest is an extremely important tool in the field of scientific visualization as it allows researchers to isolate regions based on their domain knowledge. However, the increasing size of large-scale datasets often forces users to rely on distributed computing environments which have many drawbacks in terms of interaction and convenience. Many of the current feature extraction techniques are designed around these distributed environments. The ability to overcome the memory and bandwidth limitations of desktop PCs can broaden their usability towards large-scale applications. In this work, we present a new hybrid feature extraction technique which combines GPU-accelerated clustering with the multi-resolution advantages of supervoxels in order to handle large-scale datasets on standard desktop PCs. Furthermore, this is paired with a user-driven uncertainty-based refinement approach to enhance extraction results into a desired level of detail. We demonstrate the effectiveness and interactivity of this technique using a number of application specific examples utilizing large-scale volumetric datasets.