面向大规模场景的多片元效果高效绘制

发布时间：2019-02-13 10:42

【摘要】：多片元效果具有实时透明等重要应用,它需要每个像素按深度顺序遍历对应的所有片元.深度剥离法将场景重复绘制多次来满足这个需求,故对显存带宽提出了很高的要求.该文针对大规模场景图元分布稀疏的特点,使用类体素八叉树在物体空间将场景近似剖分以减少图元读取总量.这允许场景能够按可见性顺序被分块加载、精确剖分和逐个绘制:通过光栅化对应的八叉树网格构造深度桶列表,在屏幕空间将场景块与网格布尔求交;通过为整个场景构造块的深度直方图,在剥取时利用它来避免硬件遮挡查询操作.由于为每块采取了不同的迭代次数,该文的两阶段剖分方法能够适应物体空间变化的深度复杂度.由于不依赖于面片的邻接信息,该文方法能够支持非流形网格和外存绘制.相比已有工作,该文方法在每次剥取一层时绘制效率有30%以上的提升.
[Abstract]:Multi-cell effect has many important applications such as real-time transparency and so on. It requires each pixel to traverse all the corresponding cells in depth order. In order to meet this requirement, the deep stripping method repeatedly draws the scene for many times, so the bandwidth of the video memory is very high. In this paper, based on the sparse distribution of large scale scene elements, a voxel octree is used to approximate partition the scene in object space to reduce the total amount of element reading. This allows scenes to be loaded into blocks in visibility order, accurately partitioned and drawn one by one: a depth bucket list is constructed by rasterizing the corresponding octree grid, and the scene blocks are intersected with the grid in screen space. By constructing the depth histogram of the block for the whole scene, it can be used to avoid hardware occlusion query operation. Because different iterations are used for each block, the two-stage partition method in this paper can adapt to the depth complexity of the spatial variation of the object. Because it does not depend on the adjacent information, the proposed method can support non-manifold mesh and external rendering. Compared with the previous work, the efficiency of the proposed method is improved by more than 30% when one layer is peeled.
【作者单位】：中国科学院计算技术研究所前瞻研究实验室;移动计算与新型终端北京市重点实验室;中国科学院大学;
【基金】：国家自然科学基金(61173067,61379085,61532002) 国家“八六三”高技术研究发展计划项目基金(2015AA016401)资助
【分类号】：TP391.41

【相似文献】