CHS: Small: Functional Proceduralization of 3D Geometric Models

虽然近几十年来计算能力和数字存储的可用性有了极大的提高，但创建有用的3D数字内容仍然很困难。虽然我们可以在互联网上找到大量现有的3D模型，但这些模型通常在几何上是无序的，这严重限制了它们的实用性。本研究的愿景是通过将非结构化数据转换为有组织且易于编辑的表示，促进大型复杂3D模型的半自动、可控内容创建和编辑，用于数字仿真、可视化、娱乐、教育和文化遗产。考虑到这一目标，该项目将创建数字工具，将作为输入的几何模型程序化为一组有意义的规则及其参数，从而使模型紧凑，可以以直观的方式更改，并且可以快速渲染。预计项目成果将在计算机图形学应用领域产生广泛的影响。这项工作的基础方法是找到一个过程表示（即算法、规则集和参数值），它可以在考虑其功能的同时生成给定的输入模型。该方法的一个关键新颖之处在于自动生成程序表示，并学习对象的功能和语义结构。该项目将探索在编程语言（如Python和Java Script）中生成过程表示，以及使用新的过程表示，如自适应几何图形语法或扩展来形成和分割语法。因此，该方法将通过支持生成相同对象的功能有效变体、为执行高级或低级更改公开参数以及用合理的细节完成对象来帮助创建内容。这项工作将以三个主要组成部分进行：原子化（发现输入模型的原子或终端元素）；模式化（查找并列举参数化的规则和/或参数值，这些规则和/或参数值共同用于产生输入对象）；和合成（提供算法和工具，以便有效和直观地生成输入对象或对象的合成变体）。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

While computational power and digital storage availability have increased tremendously in recent decades, the creation of useful 3D digital content is still difficult. Although we can find huge collections of existing 3D models on the Internet, these models are typically geometrically disorganized which severely limits their utility. The vision underlying this research is to facilitate semi-automatic, controllable content creation and editing of large and complex 3D models for use in digital simulation, visualization, entertainment, education, and cultural heritage, by converting unstructured data into organized and easily editable representations. With that goal in mind, this project will create digital tools that proceduralize a geometric model provided as input into a set of meaningful rules and their parameters so that the model is compact, can be altered in an intuitive way, and is fast to render. Project outcomes are expected to have broad impact in computer graphics applications spanning a wide variety of domains.The methodology underlying this work is to find a procedural representation (i.e., algorithm, rule set, and parameter values) that can generate a given input model while considering its function. A key novelty of the approach is the automatic production of a procedural representation and the learning of the function as well as the semantic structure of the object. The project will explore generation both of procedural representations in programming languages such as Python and Java Script and by using novel procedural representations such as adaptive geometric graph grammars or extensions to shape and split grammars. As a result, the methodology will help content creation by enabling the generation of functionally-valid variants of the same object, exposing parameters for performing high-level or low-level alterations, and completing the object with plausible details. The work will proceed with three main components: atomization (discovery of the atomic, or terminal, elements of the input model); patternization (finding and enumerating the parameterized rules and/or the parameter values that are collectively used to produce the input object); and synthetization (providing algorithms and tools to enable the efficient and intuitive generation of synthetic variations of the input object or objects).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.