EAGER: Minimal 3D Modeling Methodology

EAGER：最小 3D 建模方法

• 批准号: 2032770
• 负责人: Daniel Aliaga
• 金额: $ 6.5万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-07-01 至 2022-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2032770&HistoricalAwards=false
• 关键词: EAGER; Minimal; 3D; Modeling; Methodology

Modeling and design are core components of computer graphics and computer vision research and applications. Traditional modeling consists of having the designer provide either a detailed (digital) specification of the desired virtual object or sufficient photographs of a physical object to enable a multi-view stereo reconstruction, although modern GUI-based tools can help reduce to a certain extent the number of photographs required. Digital sketching tools provide an alternative mechanism for modeling objects, but even though some of these try to assist the user by completing partial sketches a notable effort is still required to achieve detailed results. This project will explore a modeling methodology that addresses the following question: What is the least we can design and still obtain a sufficiently expressive system? At one extreme digital modeling tools support high expressivity but require high design effort, while at the other extreme providing a fixed set of model templates incurs very low design effort but results in low model expressivity as well. Some recent efforts, such as the PI's sketch-to-procedural-modeling work, fall somewhere in the middle. The goal of the current research is to determine the point of optimum balance between design and expressivity, that is just enough design effort to produce a sufficiently expressive model. The focus of this multi-disciplinary work will be on computational archaeology, an interesting application where only fragmented information is available, hence success of the approach in this domain will imply broad generalizability of project outcomes to other areas as well.It has been established in the literature that only a fraction of what we perceive suffices for a person to create a mental 3D representation of an object. With this observation in mind, the project will build upon and extend the PI's existing photograph-to-3D modeling tool by adding new minimalist machine learning underpinnings applied to urban and archaeological modeling and design, to build software that requests just enough input from the user and is able to produce 3D models of sufficient completeness for the intended goal. Various ways of degrading the input detail and analyzing how the model output is affected will be explored to identify the most promising for retention and improvement of robustness. About a terabyte of imagery and point cloud data from two archaeological sites of ancient settlements on the islands of Dana and Bogsak along the southern coast of Turkey will serve as a testbed. These islands have structures built using material from local stone quarries and are a main type of urban landscape to survive from antiquity but are difficult to study due to size, complexity, terrain, and incompleteness. Nonetheless, aerial drone-based imagery and LIDAR are possible. The research will investigate how much must be specified during design to differentiate among the possible forms and their parameters to express a desired output.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.

建模和设计是计算机图形学和计算机视觉研究和应用的核心组成部分。 传统建模包括让设计者提供所需虚拟对象的详细（数字）规格或物理对象的足够照片以实现多视图立体重建，尽管现代基于 GUI 的工具可以在一定程度上帮助减少所需照片数量。 数字草图工具提供了一种用于建模对象的替代机制，但即使其中一些工具试图通过完成部分草图来帮助用户，仍然需要付出巨大的努力才能获得详细的结果。 该项目将探索一种解决以下问题的建模方法：我们至少可以设计什么并仍然获得足够表达的系统？ 在一个极端，数字建模工具支持高表现力，但需要大量的设计工作，而在另一个极端，提供一组固定的模型模板会导致非常低的设计工作，但也会导致低模型表现力。 最近的一些工作，例如 PI 的草图到程序建模工作，处于中间位置。 当前研究的目标是确定设计和表现力之间的最佳平衡点，即足够的设计努力来产生足够表现力的模型。这项多学科工作的重点将是计算考古学，这是一种有趣的应用，其中仅提供碎片信息，因此该方法在该领域的成功将意味着项目成果也可以广泛推广到其他领域。文献表明，只有我们感知到的一小部分就足以让一个人在心理上创建一个物体的 3D 表征。 考虑到这一观察结果，该项目将通过添加应用于城市和考古建模和设计的新的极简主义机器学习基础，建立并扩展 PI 现有的照片到 3D 建模工具，以构建只需要用户提供足够输入的软件并能够为预期目标生成足够完整的 3D 模型。 将探索降低输入细节并分析模型输出如何受到影响的各种方法，以确定最有希望保留和提高鲁棒性的方法。 来自土耳其南部海岸达纳岛和博格萨克岛两个古代定居点考古遗址的约 1TB 图像和点云数据将作为测试平台。 这些岛屿的结构采用当地采石场的材料建造，是古代遗留下来的主要城市景观类型，但由于规模、复杂性、地形和不完整性而难以研究。 尽管如此，基于无人机的空中图像和激光雷达是可能的。 该研究将调查在设计过程中必须指定多少内容，以区分可能的形式及其参数，以表达所需的输出。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力优点和更广泛的影响审查进行评估，被认为值得支持标准。

项目成果

Urban tree generator: spatio-temporal and generative deep learning for urban tree localization and modeling

城市树木生成器：用于城市树木定位和建模的时空和生成深度学习

• DOI: 10.1007/s00371-022-02526-x
• 发表时间: 2022-06-21
• 期刊: The Visual Computer
• 作者: A. Firoze;Bedrich Benes;Daniel G. Aliaga
• 通讯作者: Daniel G. Aliaga

Synthesis and Completion of Facades from Satellite Imagery

卫星图像的外立面合成和完成

• DOI: 10.1007/978-3-030-58536-5_34
• 发表时间: 2020-08
• 期刊: European Conference on Computer Vision
• 作者: Zhang, X.;May, C.;Aliaga, D.
• 通讯作者: Aliaga, D.

Automatic Deep Inference of Procedural Cities from Global-scale Spatial Data

从全球规模的空间数据自动深度推断程序城市

• DOI: 10.1145/3423422
• 发表时间: 2021-02
• 期刊: ACM Transactions on Spatial Algorithms and Systems
• 影响因子: 1.9
• 作者: Zhang, Xiaowei;Shehata, Aly;Benes, Bedrich;Aliaga, Daniel
• 通讯作者: Aliaga, Daniel

RFCNet: Enhancing urban segmentation using regularization, fusion, and completion

RFCNet：使用正则化、融合和补全增强城市分割

• DOI: 10.1016/j.cviu.2022.103435
• 发表时间: 2022-04-01
• 期刊: Comput. Vis. Image Underst.
• 作者: Xiaowei Zhang;Daniel G. Aliaga
• 通讯作者: Daniel G. Aliaga

Guided pluralistic building contour completion

引导多元建筑轮廓完成

• DOI: 10.1007/s00371-022-02532-z
• 发表时间: 2022-06-08
• 期刊: The Visual Computer
• 作者: Xiaowei Zhang;Wufei Ma;G. Varinlioǧlu;Nick Rauh;Liu He;Daniel G. Aliaga
• 通讯作者: Daniel G. Aliaga