Alexander Vasilevskiy

Flux maximizing geometric flows

Several geometric active contour models have been proposed for segmentation in computer vision. The essential idea is to evolve a curve (in 2D) or a surface (in 3D) under constraints from image forces so that it clings to features of interest in an intensity image. Recent variations on this theme take into account properties of enclosed regions and allow for multiple curves or surfaces to be simultaneously represented. However, it is not clear how to apply these techniques to images of low contrast elongated structures, such as those of blood vessels. To address this problem we derive the gradient flow which maximizes the rate of increase of flux of an auxiliary vector field through a curve or surface. The calculation leads to a simple and elegant interpretation which is essentially parameter free. We illustrate its advantages with level-set based segmentations of 2D and 3D MRA images of blood vessels.

3D Flux Maximizing Flows

A number of geometric active contour and surface models have been proposed for shape segmentation in the literature. The essential idea is to evolve a curve (in 2D) or a surface (in 3D) so that it clings to the features of interest in an intensity image. Several of these models have been derived, using a variational formulation, as gradient flows which minimize or maximize a particular energy functional. However, in practice these models often fail on images of low contrast or narrow structures. To address this problem we have recently proposed the idea of maximizing the rate of increase of flux of an auxiliary vector field through a curve. This has lead to an interpretation as a 2D gradient flow, which is essentially parameter free. In this paper we extend the analysis to 3D and prove that the form of the gradient flow does not change. We illustrate its potential with level-set based segmentations of blood vessels in a large 3D computed rotational angiography (CRA) data set.

A framework for reducing instruction scheduling overhead in dynamic compilers

Start-up time is a serious concern for high-availability applications such as web servers, transaction managers, and batch processes. Compilation time contributes directly to start-up costs in dynamic compilers. Up to 30% of compilation time is spent scheduling instructions in the IBM® Testarossa just-in-time compiler. In this paper, we describe a scheduling framework that reduces scheduling overhead by up to 61% with little to no degradation in throughput performance. By combining online profile-directed feedback data with information generated during register allocation, our framework identifies code regions that will benefit most from instruction scheduling. We evaluate our framework on typical client-side applications, multi-threaded server applications to production application servers on the IBM® zSeries® 990 and POWER#8482; platforms.