Visualization: Plenoptic Opacity Function for Large Date Visualization

目前，科学和医学应用程序日常产生的千兆字节到兆兆字节的数据产生了对功能强大的可视化工具的迫切需求。阻碍此类工具发展的关键问题之一是，在这些大型数据集的交互呈现过程中，可用的系统带宽与数据访问的总需求之间的差距越来越大。不幸的是，尽管不断努力加速I/O子系统，这种物理带宽的稀缺性仍然存在。作为对提高I/O性能的持续努力的补充，该项目侧重于使用一种独特的、粗粒度的可见性筛选方法，智能地减少对数据访问的需求。可见性剔除旨在通过关注从特定视图可见的数据部分来消除不必要的可视化计算，在可视化管道的早期阶段剔除遮挡部分，这可能是实质性的。尽管可见性剔除之前已经作为渲染过程的副产品进行了研究，但它还没有作为一种独立的技术来解决带宽瓶颈。为此，我们在这里提出的项目将专注于发现和开发新的通用、可扩展和高效的可见性剔除算法。我们的关键概念是通过预计算对元不透明度进行编码。在任何体积中，每个体素都有一个预先确定的不透明度值。在渲染过程中，一个体素序列在深度上被合成，以形成一个与视图相关的不透明度值，称为元不透明度，用于每个射线段。我们假设存在一种方法，通过一次性的预计算，对一系列传递函数下的一系列视图的凸形体块的元不透明度进行保守而有效的编码。我们称这种编码为全光不透明度函数（POF）。使用POF进行保守的运行时不透明度估计，可以实现体绘制中可见性剔除的极大加速，同时可以在同一族内动态修改传递函数。在这里，我们验证了这一假设，并将POF应用于大规模并行的体绘制，以及核外可视化。

The gigabytes to terabytes of data now being produced by scientific and medical applications on a routine basis create a pressing demand for capable visualization tools. One of the key problems hindering the development of such tools is the growing disparity between the available system bandwidth and the total need of data access during interactive rendering of these large data sets. Unfortunately, such scarcity of physical bandwidth persists in spite of repeated efforts to accelerate I/O subsystems. Complementary to those continued efforts in increasing I/O performance, this project focuses on intelligently decreasing the need of data access using a unique, coarse-grained method of visibility culling. Visibility culling aims at eliminating unnecessary visualization computation by focusing on parts of the data that are visible from a specific view, culling occluded portions, which may be substantial, in the early stages of the visualization pipeline. Although visibility culling has previously been done as a by-product of the rendering process, it has yet to be explored as an independent technique for attacking the bandwidth bottleneck. For that purpose, the project we propose here will focus on the discovery and development of new visibility culling algorithms that are generic, scalable and efficient. Our key concept is to encode meta-opacity via precomputation. In any volume each voxel has a predetermined value of opacity. During rendering, a sequence of voxels is composited in depth order to form a view-dependent opacity value, called meta-opacity, for each ray segment. We hypothesize that there exists a way to conservatively and efficiently encode meta-opacity of a convex-shaped volume block for a range of views under a family of transfer functions via a one-time pre-computation. We call such encoding the Plenoptic Opacity Function (POF). Using POF for conservative run-time opacity estimation, great accelerations by visibility culling in volume rendering can be accomplished while the transfer function can be modified within the same family on the fly. Here we test this hypothesis and apply POF to volume rendering in large-scale parallelism, as well as out-of-core visualization.