Procedural Creation and Real-time Display of Realistic Complex Scenes

真实图像合成的关键要素需要处理场景几何形状的表示，其单独/组合纹理和反射率属性，以及发射和入射光如何穿过它。高质量的离线渲染，通常基于蒙特卡罗路径跟踪，很好地解决了相关问题，但代价是数千小时的计算和千兆字节的内存空间。然而，许多需要实时逼真显示的应用必须处理这些元素，并在不同的尺度上进行某种形式的预滤波。我们引入了用于实时渲染的分层纹理/表面/体方法。它们是在规则壳图和千兆像素的变化基础上构建的，在体积元素中提供新的创新近似，以实现对外观至关重要的视觉现象，同时保证每个像素或每个光线的单一结构查询。本研究计划深入研究这些表示，并将它们与当今的GPU设施很好地集成在一起。这些方法和结构也与艺术家和科学家的直觉控制相结合，以产生可信和连贯的复杂世界。

The key elements to realistic image synthesis need to handle the representation of scene geometry, its individual/combined textures and reflectance properties, and how emitted and incident light travels through it. High-quality offline rendering, often based on Monte Carlo Path Tracing, solves well the related problems, but at the cost of thousands of hours of computations and gigabytes of memory space. However, many applications requiring real-time realistic display must handle these elements all together with some form of pre-filtering appearances at different scales. We introduce hierarchical texture/surface/volume methods for real-time rendering. They are constructed on variations of regular shell maps and gigavoxels, providing new innovative approximations within volume elements to visual phenomena critical to appearance, while guaranteeing a one-structure query per pixel or per ray. This research program dives into these representations and integrates them well with today's GPU facilities. These methods and structures are also combined with intuitive controls for artists and scientists to generate credible and coherent complex worlds.