Your browser doesn't support the features required by impress.js, so you are presented with a simplified version of this presentation.

For the best experience please use the latest Chrome, Safari or Firefox browser.

smart urban form transformation database :An machine learning and urban morphology Based approach for AUTomatic generation of Synergetic urban plan


How to 自动化 GENERATE new urban design in the city context in a way that the new urban form improve the entire urban Synergy and its sustainability is the main question to be discussed in this paper.

Study of urban form is an important area of research in urban planning/design that contributes to our understanding of how cities function and evolve. However, classical approaches are based on very limited observations and inconsistent methods.

快速的城市生成和预测的问题研究,一方面对于城市形冭学量化的研究算法局限于算法和生成的城市往往只考虑单独的元素而不是城市语境 ,另一个方面新的ai技术生成城市,人们难以有具体指标去控制最后的成果

This paper proposes deeplearning based approach to generate considering urban contexts, rather than individually analyzing the urban elements. We created data including the morphological characteristics of the buildings and their surrounding urban spaces. By 训练the data, statistical techniques identified them pathological features and types to 交互式的 generate city

以米兰为例 城市形态转型。 How to 快速的生成urban design in the city context in a way that the new elements improve the entire neighborhood 舒适 performance and its sustainability is the main question to be discussed in this paper

we trained a deep convolutional auto-encoder, that automatically learns the hierarchical structures of urban forms and represents them via dense and comparable vectors.

以米兰做运河地区做实验。自动的在osm上找到了2000个满足标准的样本。gaugan 模型和分布式训练,分别训练了两个机器学习数据库,一个用于路网和形态边界的生成,一个用于建筑元素的生成。设计师布局控制形态的关键因子地块开口和内部空地, 就可以快速的生成多种城市的结构和建筑3d精细模型排布来寻找城市设计的解决方案来解决 米兰大量的量化形态学研究和大量样本研究,快速的

以 ,功能混合度,地块密度,交通可达性,平均高度,多各方面来验证城市生成工具的适用性和城市结构预测能力.

发现同比于原来的地块 ,ai生成的方案很好的解决了城市密度和功能混合度问题,相比于传统算法更多的适应性和方案可能性,但是对于交通能力和原有地块相比类似。

This paper reports on the analytical potential of machine learning methods for urban design generation based on

Therefore, this research aims to apply GAN in creating urban design plans, helping designers automatically generate the predicted details of buildings configuration with a given condition of cities.

The proposed methodology was tested on the French Riviera and outputs show a moderate predictive capacity (i.e., adjusted R2 = 0.75) and insightful explanations on the nuanced relationships between selected features of the urban environment and street values. These

A novel index termed Piping Coverage Rate was jointly proposed to evaluate the obtained results.

The system produces the output within 45 seconds, which is drastically faster than the conventional manual workflow.

can we design by learning from data?

And can we integrate physical and non-physical elements of spaces in one coherent informational space?


Synergetic urban planing (sup)

量化城市形态 for Synergetic urban planing (sup)

Study of synergetic urban planing(sup) is an important area of research in urban morphology form that contributes to A SUSTAINABLE URBAN transformation.

从简· 雅各布斯(Jane Jacobs)关于城市多样性的讨论 [6]
,到杨 · 盖尔关于街道空间活力的建议 [7],以及随后诸多城市设计研究者和实践者 的进一步讨论,都提出了sup 城市的理论

然 而在相关理论日益增多的同时,城市空间活力营造依然被认为是一 个难以明确界定而且依赖于设计师经验和直觉的过程 [8,9]

in the traditional way, designers spend a lot of time working on sketching for different characteristics of urban design in the initial stage for a sup

Historically, urban morphologist study cities based on qualitative approaches and personal observations and limited to few famous cities or even one specific location1–3.(2017Moosavi, Vahid)

morphology 否定

尽管SUP 很重要 classical approaches are based on very limited observations and inconsistent methods.

the static and sectoral approaches cannot fully fulfill such complex and dynamic requirements.

主要的挑战是没有大量的量化urban 形态学form data aisist urban design ,生成考虑城市生成问题只有人工有限样本的城市形态学的定量单独分析

整体性方法 for sup

Within the domain of urban studies, urban morphology is the classical study of urban forms and their underlying formation processes and forces over time

urban planning require detailed information about the functional, morphological and socio-economic structure of the built environment. The building stock directly affects urban structure, e.g. urban form, density of housing and distribution of population. (2013Robert Hecht*)


Much research has been done in different urban sciences with the consideration of the city as a complex system (CS) (2011Manesh, Shahrooz Vahabzadeh
Tadi, Massimo)

This system can comprise different kinds of urban artifacts (Rossi 1982) - human- made objects such as buildings, infrastructures, pub- lic space, etc

(Conzen School)的相关理论,城市形态的关键要素包括城 镇平面(街道、地块和建筑),建筑组构以及土地利用等 [12,13]

urban morphology character

量化urban morphology character as a key for Synergetic urban plan generation 有很多相关的理论

绝大多数城市空间活力的营造原则可以被归纳为三个关 键的城市形态要素:良好的街道可达性、适宜的建设强度与 建筑形态、足够的功能混合度(表 1)。

urban morphology character

however ,only the feature metrics of individual urban artifacts and hence fail to capture contextual and relational aspects of the city. 诸多城市形态学关注的都是城市单独元素而不是城市的完整系统。

how to 应用整体性的方法 处理大量的urban morphology character and morphology structure assist Synergetic urban plan generation in a fast way 是一个难题

In this paper we 具体的define sup 方式 as the physical conditions of a range 城市的结构 密度 形式 功能 高度 等形态学讯息 综合形态学特征的综合性考虑

deeplearning for sup

从评估到生成一个sup 城市,计算机扮演重要角色尤其是ml

computers are playing an increasingly important role in the design process. Man phases in design once carried out by hand are nowadays automated (Bolognini et al. 2011).

Automatic generation

In the context of urban design, automation enables the fast generation and exploration of different design options using computational tools so that designers can acquire a comprehensive overview and understanding of different design choices. 2018 Miao, Yufan; Koenig, Reinhard; Knecht et al.


urban forms have complex and hierarchical structures, composed of building layouts, street segments, neighborhood layouts and main street networks. Therefore, while network based analytics is the common approach in quantitative urban studies, as we will show later considering the inherent complexity of urban patterns, the use of the so-called representation learning and deep learning algorithms1

deep learning techniques on the other have opened up new possibilities to automatically investigate urban forms at the global scale .Moosavi, Vahid2017

(2013Hecht, Robert)automatic derivation of urban structures types by using (ml) with focus on residential areas.

data-driven urban analysis based on diagrammatic images describing each building in a city in relation to its immediate urban context.

(Moosavi, Vahid2017)trained a deep convolutional auto-encoder, that automatically learns the hierarchical structures of urban forms

considering their overall spatial structure and other factors such as orientation, graphical structure, and density and partial deformations

Various computational methods in studies of urban morphology have been employed to define and distinguish patterns of a city, such as the statistical model (Colaninno et al. 2011), space syntax (Alexander 1965), convolutional neural networks (Al- bert et al 2017), among others.

morphology /automatic

However, their research focus on analysis smilar and 聚类大量的城市形态特征 还没有 根据具体的城市语境生成建筑排布,形成tool that asisst designers 处理大量的城市样本形态和快速交互的响应式辅助来生成一个城市设计

relate work

Through the retrieval of existing articles, the following three used related meth-ods of computer vision achieving substantial results in the field of building plan generation. The first two articles are mainly focused on the generation of indoor floor plans, while the last article is aimed at general layout. The results all adopt a similar method: first convert building layout into bitmaps, and then use image to image algorithm to train the deep learning model.

In 2018, Hao Zheng trained 100 samples using pix2pixHD to realize the recog-nition and generation of architectural layouts. First, it trains with indoor layout as label and function color map as result; at the same time, it reserves training to achieve the mutual map of the indoor layout and the function color map; The visualization of the training process proves that pix2pixHD has similarities with human cognition. The author mentioned that results couldn’t match the irregular boundary well that may be due to samples which have irregular boundary are few.

Stanislas Chaillou produced the indoor layout generator program ArchiGan in 2019. Using GANs and step training, three models can realize from site conditions to building outline to function color map to indoor layout. When inputing the site condition, it can output the indoor layout. In the third step, it conducts separate training the furniture layout of each type of room through different colors.Step training is to control the single steps of machine learning, to realize the human intervention and ensure generation quality.It trained four architectural styles to achieve style transfer to be selected according to pros and cons when designing.

The schgan produced by Yubo Liu in 2019 can automatically generate a functional layout of elementary school based on road and contour conditions. According to the user evaluation mechanism set by it, the plan generated by AI is higher than the original plan.Specific experimental method was not disclosed, but the results prove that the deep learning can achieve generation of general layouts.

apply GAN in creating urban design plans, helping designers automatically generate the predicted details of buildings configuration with a given condition of cities.(2020JIAQI SHEN1, CHUAN LIU2)

The above articles have the following questions: (1) Due to the limits of algorithm and the number of samples, the generated results have a lot of noise, and the visual recognition is greatly affected;(2) The above articles exist that the direction of elements in the results can only be orthogonal, and cannot change according to site condition. This experiment proves that reasons include not only the above-mentioned oblique arrangement, but also the limitations of pix2pix itself; (3) The above researches can not realize diversity generated results with a single input, which is impossible to provide designers with multiple possibilities.多样性 4 输入因素无法控制具体的城市形态解决城市协同性问题

goal for sup

This paper proposes an interactive way to automatic generate sup by deep learning model and morphological rules considering 大量的 urban contexts in a整体性的方式 , rather than 单独城市某个因素

We call this method Computational Synetic Urban planing (CsUP). It aims to support urban designs through releasing people 耗时 手工 慢速的 work on drawing and analysing complex urban 形态学 form and strutur. 减少了因为人们的感知和反复修改带来的巨大重复工作量。

csup 收集 data form osm data platform including the morphological characteristics of the buildings and their surrounding urban structure types 。 By 训练the data baed gausgan model that automatically learns the hierarchical structures of urban forms and fast provide the urban plan based on 数据库 in a few second

The final goal of this research is to build an interactive tool that takes 形态学设计规则 (empty void boundary 方向性)as input, 经过computer trained data库 处理 , output image data showing 综合的 urban planning with 预测 details of buildings configuration

designers to interact with the generated designs not only 机器学习形态学参数 inputs but also geometrically by changing geometric constraints.

The following sections describe our method in technical detail and illustrate it through a case study in the city of milan.

We validate our result 在 多样性 生成时长 urban 协同性 等方面

多个综合urban index was jointly proposed to evaluate the obtained results compared with 现在的方案。 and reflect on some limitations and avenues for future work

case study

The research presented here evolved within the context of Milan Naviglio grande urban form transformation project as a studio 教学at Politecnico di Milan. The project focuses on GENERATE new urban design in the city context in a way that the new urban form improve the entire urban Synergy and its sustainability

难点 1 传统方式需要人工find and analysis 大量的形态学form 来aisist 地块的urban design 量化 urban form 形态学 而不是感知 来符合城市语境

2 in urban process different stage , 很难整体性考虑城市而不是urban element which 会导致大量的人力和cost 在方案各个阶段和各个角色的协同性上。

在这种情况下 Our tool developed to enhance the urban design process with two main characteristics:
automated的 大量样本学习 ,和 交互式的协同的planigng

to generate spatial configurations that fulfil all restrictions, which are primarily concept stage: tool to develop a new 输入根据边界和方向性,generate路网plan and 预测地块形态特性( 边界的开放void 地块内部的empty属性) for 邻里尺度fit 规划 requirments based on ai 提供and designers 修正

2in项目中期 地块之内的尺度. develope model with input形态学特征时(用地边界 路网开口 empty),机器能快速的自动化生成城市building arrangement and detail spatial configurations (funtion density accessbility )for building 的分布和设计细节优化。

3 for 项目后期 矢量化得到 ,多个urban solution 3d model with visualazation performance (可达性 功能混合度) it saves manpower costs and improves the quality of the design solutions by making it possible to test and evaluate different variants rapidly


数据收集 城市data 城市学习数据讯息样本

训练的图集的data主要分为两个模型。 which 用来解决不同的sup 的生成 1城市路网和 urban 网格 character的预测ban open border emty 2城市building arrangement 和building character 的预测 . 第一个数据集的训练结果可以作为第二个的输入,也可以单独使用gan 对他们培训以得到不同的尺度和需求的结果。

在houdini中我们选取了milan的osm开源的数据which 记录着建筑的讯息


使用768m x 768m的正方形进行选取,一次得到地块数>4个则作为样本,一共选取了801个


绿色通道(见二级模型input):open,voild的反向,制作方法为在边界上采样点向内发射ray撞击建筑边界,撞击点到发射起始点的距离超过20m即open点,20m以内即voild。删除voild部分,得到open小短线再次向内offset 6m,使相邻每个地块的open不重叠。
蓝色通道:建筑边界向外offset 12m,场地边界向内offset 12m剩余部分即empty。


再计算,每个地块的建筑功能混合度>0.3(建筑功能数 / 建筑个数)以及建筑密度>0.3则被选取。作为样本,一共选择了1017个样本

蓝色通道:建筑边界向外offset 12m,场地边界向内offset 12m剩余部分即empty。

绿色通道:代表建筑高度信息,g值 = clamp(建筑高度 / 25.0, low=0, height=1) * 0.8 + 0.2)
蓝色通道:代表建筑功能信息,b值:Accommodation = 0/6,
Commercial = 1/6, Religious = 2/6, Civic = 3/6, Agricultural = 4/6, Facility = 5/6, others = 6/6

generation of GAN machine learning modular


uses the GauGAN algorithm to complete the model architecture and us-es step-by-step training (outline-> buildings and buildings-> roadmap)

GauGAN neural network architecture
GauGAN is an image translation algorithm published by Nvidia Lab in 2019 that can achive multi-modal synthesis. This experiment is implemented according to the paper, including VAE, which is used to achieve style guide multi-modal synthesis. The implementation of GauGAN is shown in Fig. 1.4.3(1).

step training.

step training is to map the contour graph to the building layout graph first, and then map the build-ing layout graph to the road network graph. When testing, matrix multiplication is applied on the obtained building layout and road network to obtain the final re-sult.


test moudular

test 流程流程:
3.然后分别矢量化(potracelib C++)三个通道为多边形,并将用路网划分出地块,将open以及empty记录到每个地块上。
5.分布矢量化建筑边界用于区分每个建筑的区域,用得到的每个建筑的区域在位图结果中计算该区域的平均Green值以及blue值,建筑高度 = (G - 0.2)/0.8 * 25(米),功能与blue通道值对应关系见二级模型output制作

Vectorization and 3D procedural modeling

Models trained by the step training generates ??? The bitmap results are vectorized and merged, and then 3d procedural modeling and visualization are performed.
什么技术? 怎么做的矢量转化 高度等等

流程of case study in milan

the site located in milan ,figure x shows the
urban generation by ai model

1 requirements

for a better tranformation of urban form ,3 indicators are needed

满足sup的 spatial configurations which are primarily:
2efficient use of the available 居住unit on the site with 更高密度
3 让open space 高可达性的 communities by arranging households in a smart way

2Generation of Street Networks and 形态学特征

Using the methods described below, we generate basic urban layouts as depicted in figure 11. For the generation , the designer need only specify

1 the border and of the planning area,
2 the 入口 of entry,
3 the 方向向量

by inputing and ajusting 上述的形态学 character figure xx,plan showing a basic network showing open border

和 上面的比更多的open 和更分散的公共emty

Generation of buildings and 空间配置

5.分布矢量化建筑边界用于区分每个建筑的区域,用得到的每个建筑的区域在位图结果中计算该区域的平均Green值以及blue值,建筑高度 = (G - 0.2)/0.8 * 25(米),功能与blue通道值对应关系见二级模型output制作

Using then 二级model generated urban building layouts如图 ,showing not only building arrangement as well as 具体建筑
数据预测 (高度 function density accessibility)

designers can change 不单是 输入参数 open emty rotation 还有 geomtry of urban element

Exploring Design Variants

3d 模型生成 结果

vacant land = sum(blue areas)(就是empty面积)
open border = sum(green lines)(就是open总长度)
accessibility = sum(node_count_r400m ^ 1.2 / (total_depth_r400m + 2))(每条道路的,空间句法路网活跃度Nach_r400m,之和)

building density = 建筑面积 / 地块面积
function diversity = 每个地块的,建筑功能数 / 建筑个数,的平均
average height = 建筑平均高度
accessibility = 每个建筑到vacant land的最短距离,的平均值


3d 矢量化 model of milan 地块

最终的方案 camparation with original

图比 密度和几个基础特性

比各个空间到open 植入空间的emty 的可达性

result Discussion







it reduces the time that urban designers spend exploring different variants of urban layouts; it saves manpower costs and improves the quality of the design solutions by making it possible to test and evaluate different variants rapidly. It also makes the design process explicit and traceable and designs scalable and reproducible. Collaboration and communication platforms can additionally help to connect, engage and better facilitate the discussion of alternative designs and their respective trade-offs, and to involve different stakeholders



Or you can simply list your references here:

  1. some ref
  2. some other ref. Numbering fixes itself automatically.
  3. A third ref.


  1. Xuan Yang (2011) The Effect of Neighborhood Mode in Modern Residential District Plan-ning. Chinese & Oversea Architecture, 94-95
  2. Jia Rongxiang, Sun Ying (2012) Cultural Characteristics and Existence Value of Buildings in Beijing Million Village. Journal of Beijing Institute of Civil Engineer and Architecture, 76-80
  3. Li Hao (2014) The introduction of Soviet planning theory to China. Beijing Planning and Construction, 165-168
  4. Dan Hendrycks, Kevin Gimpel (2016) GAUSSIAN ERROR LINEAR UNITS (GELUS), University of California, Berkeley, Toyota Technological Institute at Chicago
  5. Weixin Huang, Hao Zheng (2018) Architectural Drawings Recognition and Generation through Machine Learning, Tsinghua University, University of Pennsylvania
  6. Stanislas Chaillou (2019) AI Architecture Towards a New Approach
  7. Phillip Isola, Jun-Yan Zhu, Tinghui Zhou Alexei A. Efros (2016) Image-to-Image Transla-tion with Conditional Adversarial Networks, Berkeley AI Research (BAIR) Laboratory, UC Berkeley
  8. Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, Bryan Catanzaro (2017) High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs
  9. Taesung Park, Ming-Yu Liu, Ting-Chun Wang, Jun-Yan Zhu (2019) Semantic Image Syn-thesis with Spatially-Adaptive Normalization
  10. Yubo Liu, Qiaoming Deng, Linyu Liang (2020) SchGAN:Primary school campus layout generation, intelligent assistant for architectural design