Institute for Data Analysis and Visualization

We present an out-of-core, point-based approach for interactive rendering of very large volumetric datasets. Our approach is based on the assumption that the density of voxels with the same function-value in large discretized volumetric scalar fields is high enough to be used to render contour and volume approximations using points to represent the voxels. This approach allows us to visualize isovalue-structures in high-resolution datasets at full resolution and interactive frame rates. In a pre-processing step, we sort the voxels by function-value and store them in a file together with a look-up table for later interactive retrieval. The displayed voxelsets can then be changed in real time by determining their locations in the file and loading them into memory. As we store position, and not function-value, the volumetric dimension of a dataset to be handled by our approach is limited by three factors: the number of points that can be rendered to achieve a sufficient frame rate, the number of bits used to store the position data, and the maximum file-size supported by the operating system. Depending on the spatial distribution of the voxels among the function-values selected, the result is either one or multiple contours or ``isoclouds''.