Search Results

Scientists and engineers are making increasingly use of hp-adaptive discretization methods to compute simulations. While techniques for isocontouring the high-order data generated by these methods have started to appear, they typically do not facilitate interactive data exploration. This work presents a novel interactive approach for approximate isocontouring of high-order data. The method is based on a two-phase hybrid rendering algorithm. In the first phase, coarsely seeded particles are guided by the gradient of the field for obtaining an initial sampling of the isosurface in object space. The second phase performs ray casting in the image space neighborhood of the initial samples. Since the neighborhood is small, the initial guesses tend to be close to the isosurface, leading to accelerated root finding and thus efficient rendering. The object space phase affects the density of the coarse samples on the isosurface, which can lead to holes in the final rendering and overdraw. Thus, we also propose a heuristic, based on dynamical systems theory, that adapts the neighborhood of the seeds in order to obtain a better coverage of the surface. Results for datasets from computational fluid dynamics are shown and performance measurements for our GPU implementation are given.