Cumincad : CUMINCAD Papers : Search Results

CumInCAD is a Cumulative Index about publications in Computer Aided Architectural Design
supported by the sibling associations ACADIA, CAADRIA, eCAADe, SIGraDi, ASCAAD and CAAD futures

PDF papers
References

Hits 1 to 20 of 613

Reformat results as: short short into frame detailed detailed into frame

_id	caadria2021_039
id	caadria2021_039
authors	Chen, Jielin, Stouffs, Rudi and Biljecki, Filip
year	2021
title	Hierarchical (multi-label) architectural image recognition and classification
doi	https://doi.org/10.52842/conf.caadria.2021.1.161
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 161-170
summary	The task of architectural image recognition for both architectural functionality and style remains an open challenge. In addition, the paucity of well-organized, large-scale architectural image datasets with specific consideration for the domain of architectural design research has hindered the exploration of these challenging tasks. Drawing upon images from the professional architectural website ArchdailyÂ®, and leveraging state-of-the-art deep-learning-based classification models, we explore a hierarchical multi-label classification model as a potential baseline for the task of architectural image classification. The resulting model showcases the potential for innovative architectural discipline-related analyses and demonstrates some heuristic insights for visual feature extraction pertaining to both architectural functionality and architectural style.
keywords	image recognition; hierarchical classification; multi-label classification; architectural functionality; style
series	CAADRIA
email
last changed	2022/06/07 07:55

_id	ecaade2021_203
id	ecaade2021_203
authors	Arora, Hardik, Bielski, Jessica, Eisenstadt, Viktor, Langenhan, Christoph, Ziegler, Christoph, Althoff, Klaus-Dieter and Dengel, Andreas
year	2021
title	Consistency Checker - An automatic constraint-based evaluator for housing spatial configurations
doi	https://doi.org/10.52842/conf.ecaade.2021.2.351
source	Stojakovic, V and Tepavcevic, B (eds.), Towards a new, configurable architecture - Proceedings of the 39th eCAADe Conference - Volume 2, University of Novi Sad, Novi Sad, Serbia, 8-10 September 2021, pp. 351-358
summary	The gradual rise of artificial intelligence (AI) and its increasing visibility among many research disciplines affected Computer-Aided Architectural Design (CAAD). Architectural deep learning (DL) approaches are being developed and published on a regular basis, such as retrieval (Sharma et al. 2017) or design style manipulation (Newton 2019; Silvestre et al. 2016). However, there seems to be no method to evaluate highly constrained spatial configurations for specific architectural domains (such as housing or office buildings) based on basic architectural principles and everyday practices. This paper introduces an automatic constraint-based consistency checker to evaluate the coherency of semantic spatial configurations of housing construction using a small set of design principles to evaluate our DL approaches. The consistency checker informs about the overall performance of a spatial configuration followed by whether it is open/closed and the constraints it didn't satisfy. This paper deals with the relation of spaces processed as mathematically formalized graphs contrary to existing model checking software like Solibri.
keywords	model checking, building information modeling, deep learning, data quality
series	eCAADe
email
last changed	2022/06/07 07:54

_id	caadria2021_038
id	caadria2021_038
authors	Chen, Jielin and Stouffs, Rudi
year	2021
title	From Exploration to Interpretation - Adopting Deep Representation Learning Models to Latent Space Interpretation of Architectural Design Alternatives
doi	https://doi.org/10.52842/conf.caadria.2021.1.131
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 131-140
summary	An informative interpretation of the hyper-dimensional design solution space can potentially enhance the cognitive capacity of designers with respect to both conventional design practice and the research domain of computational-aided generative design. However, the hitherto research of design space exploration has had limited focus on the interpretation of the hyper solution space per se due to the knowledge gap pertaining to representation and generation. Representation learning techniques, as a core paradigm in the statistically empowered domain of machine learning, possess the capability of extracting a convoluted probabilistic distribution of hyperspace with latent features from unorganized data sources in a generalized manner, which can be an intuitive modus operandi for a structural interpretation of the intricate latent design solution space and benefit the challenging task of architectural design exploration. We examine and demonstrate the potential capabilities of representation learning techniques for the interpretation of latent architectural design solution space with consideration of disentanglement and diversity.
keywords	Design space exploration; latent space interpretation; representation learning; deep generative modelling; generative architectural design
series	CAADRIA
email
last changed	2022/06/07 07:55

_id	caadria2021_086
id	caadria2021_086
authors	Eisenstadt, Viktor, Arora, Hardik, Ziegler, Christoph, Bielski, Jessica, Langenhan, Christoph, Althoff, Klaus-Dieter and Dengel, Andreas
year	2021
title	Exploring optimal ways to represent topological and spatial features of building designs in deep learning methods and applications for architecture
doi	https://doi.org/10.52842/conf.caadria.2021.1.191
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 191-200
summary	The main aim of this research is to harness deep learning techniques to support architectural design problems in early design phases, for example, to enable auto-completion of unfinished designs. For this purpose, we investigate the possibilities offered by established deep learning libraries such as TensorFlow. In this paper, we address a core challenge that arises, namely the transformation of semantic building information into a tensor format that can be processed by the libraries. Specifically, we address the representation of information about room types of a building and type of connection between the respective rooms. We develop and discuss five formats. Results of an initial evaluation based on a classification task show that all formats are suitable for training deep learning networks. However, a clear winner could be determined as well, for which a maximum value of 98% for validation accuracy could be achieved.
keywords	deep learning; spatial configuration; data representation; semantic building fingerprint
series	CAADRIA
email
last changed	2022/06/07 07:55

_id	caadria2021_305
id	caadria2021_305
authors	Keshavarzi, Mohammad, Afolabi, Oladapo, Caldas, Luisa, Yang, Allen Y. and Zakhor, Avideh
year	2021
title	GenScan: A Generative Method for Populating Parametric 3D Scan Datasets
doi	https://doi.org/10.52842/conf.caadria.2021.1.091
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 91-100
summary	The availability of rich 3D datasets corresponding to the geometrical complexity of the built environments is considered an ongoing challenge for 3D deep learning methodologies. To address this challenge, we introduce GenScan, a generative system that populates synthetic 3D scan datasets in a parametric fashion. The system takes an existing captured 3D scan as an input and outputs alternative variations of the building layout including walls, doors, and furniture with corresponding textures. GenScan is fully automated system that can also be manually controlled by a user through an assigned user interface. Our proposed system utilizes a combination of a hybrid deep neural network and a parametrizer module to extract and transform elements of a given 3D scan. GenScan takes advantage of style transfer techniques to generate new textures for the generated scenes. We believe our system would facilitate data augmentation to expand the currently limited 3D geometry datasets commonly used in 3D computer vision, generative design and general 3D deep learning tasks.
keywords	Computational Geometry; Generative Modeling; 3D Manipulation; Texture Synthesis
series	CAADRIA
email
last changed	2022/06/07 07:52

_id	ecaade2021_038
id	ecaade2021_038
authors	Nakabayashi, Mizuki, Fukuda, Tomohiro and Yabuki, Nobuyoshi
year	2021
title	Mixed Reality Landscape Visualization Method with Automatic Discrimination Process for Dynamic Occlusion Handling Using Instance Segmentation
doi	https://doi.org/10.52842/conf.ecaade.2021.2.539
source	Stojakovic, V and Tepavcevic, B (eds.), Towards a new, configurable architecture - Proceedings of the 39th eCAADe Conference - Volume 2, University of Novi Sad, Novi Sad, Serbia, 8-10 September 2021, pp. 539-546
summary	Mixed reality (MR), which blends real and virtual worlds, has attracted attention as a visualization method in landscape design. MR-based landscape visualization enables stakeholders to examine landscape changes at actual scale in real-time at the actual project site. One challenge in MR-based landscape visualization is occlusion, which occurs when virtual objects obscure physical objects that are in the foreground. Previous research proposed an MR-based landscape visualization method with dynamic occlusion by using semantic segmentation of deep learning. However, this method has two problems. The first is that the same kind of objects that are grouped into one or overlapped types are classified as the same object, and the other is that the foreground objects have to be defined in pre-processing. In this study, we developed a system for large-scale MR landscape visualization that enables the recognition of each physical object individually using instance segmentation, and it is possible to accurately represent the positional relationship by comparing the coordinate information of the 3D virtual model and all physical objects.
keywords	landscape visualization; mixed reality; instance segmentation; dynamic occlusion handling; deep learning
series	eCAADe
email
last changed	2022/06/07 07:59

_id	caadria2021_080
id	caadria2021_080
authors	Yang, Xuyou and Xu, Weishun
year	2021
title	A Tool for Searching Active Bending Bamboo Strips in Construction via Deep Learning
doi	https://doi.org/10.52842/conf.caadria.2021.1.463
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 463-472
summary	As an alternative material for construction, the structural use of bamboo in architecture is commonly associated with active bending. However, as natural material, the deformation of unprocessed bamboo strips is affected by the distribution of nodes, whose impact on deformation is difficult to precisely programme for each individual case and thus often causes discrepancies between generic digital simulation and construction. This research proposes a tool for searching active bending bamboo strips via deep leaning based on a multi-task neural network. The tool is able to predict both the number and locations of nodes suggested on bamboo strips according to a target curve as tool input. By approximating the prediction, users can find a strip that is most likely to deform into the desired geometry.
keywords	neural network; active bending; neural architecture search (NAS); bamboo; material behaviour
series	CAADRIA
email
last changed	2022/06/07 07:57

_id	acadia21_470
id	acadia21_470
authors	£ochnicki, Grzegorz; Kalousdian, Nicolas Kubail; Leder, Samuel; Maierhofer, Mathias; Wood, Dylan; Menges, Achim
year	2021
title	Co-Designing Material-Robot Construction Behaviors
doi	https://doi.org/10.52842/conf.acadia.2021.470
source	ACADIA 2021: Realignments: Toward Critical Computation [Proceedings of the 41st Annual Conference of the Association of Computer Aided Design in Architecture (ACADIA) ISBN 979-8-986-08056-7]. Online and Global. 3-6 November 2021. edited by B. Bogosian, K. Dörfler, B. Farahi, J. Garcia del Castillo y López, J. Grant, V. Noel, S. Parascho, and J. Scott. 470-479.
summary	This paper presents research on designing distributed, robotic construction systems in which robots are taught construction behaviors relative to the elastic bending of natural building materials. Using this behavioral relationship as a driver, the robotic system is developed to deal with the unpredictability of natural materials in construction and further to engage their dynamic characteristics as methods of locomotion and manipulation during the assembly of actively bent structures. Such an approach has the potential to unlock robotic building practice with rapid-renewable materials, whose short crop cycles and small carbon footprints make them particularly important inroads to sustainable construction. The research is conducted through an initial case study in which a mobile robot learns a control policy for elastically bending bamboo bundles into designed configurations using deep reinforcement learning algorithms. This policy is utilized in the process of designing relevant structures, and for the in-situ assembly of these designs. These concepts are further investigated through the co-design and physical prototyping of a mobile robot and the construction of bundled bamboo structures. This research demonstrates a shift from an approach of absolute control and predictability to behavior-based methods of assembly. With this, materials and processes that are often considered too labor-intensive or unpredictable can be reintroduced. This reintroduction leads to new insights in architectural design and construction, where design outcome is uniquely tied to the building material and its assembly logic. This highly material-driven approach sets the stage for developing an effective, sustainable, light-touch method of building using natural materials.
series	ACADIA
type	paper
email
last changed	2023/10/22 12:06

_id	ascaad2021_074
id	ascaad2021_074
authors	Belkaid, Alia; Abdelkader Ben Saci, Ines Hassoumi
year	2021
title	Human-Computer Interaction for Urban Rules Optimization
source	Abdelmohsen, S, El-Khouly, T, Mallasi, Z and Bennadji, A (eds.), Architecture in the Age of Disruptive Technologies: Transformations and Challenges [9th ASCAAD Conference Proceedings ISBN 978-1-907349-20-1] Cairo (Egypt) [Virtual Conference] 2-4 March 2021, pp. 603-613
summary	Faced with the complexity of manual and intuitive management of urban rules in architectural and urban design, this paper offers a collaborative and digital human-computer approach. It aims to have an Authorized Bounding Volume (ABV) which uses the best target values of urban rules. It is a distributed constraint optimization problem. The ABV Generative Model uses multi-agent systems. It offers an intelligent system of urban morphology able to transform the urban rules, on a given plot, into a morphological delimitation permitted by the planning regulations of a city. The overall functioning of this system is based on two approaches: construction and supervision. The first is conducted entirely by the machine and the second requires the intervention of the designer to collaborate with the machine. The morphological translation of urban rules is sometimes contradictory and may require additional external relevance to urban rules. Designer arbitration assists the artificial intelligence in accomplishing this task and solving the problem. The Human-Computer collaboration is achieved at the appropriate time and relies on the degree of constraint satisfaction with fitness function. The resolution of the distributed constraint optimization problem is not limited to an automatic generation of urban rules, but involves also the production of multiple optimal-ABV conditioned both by urban constraints as well as relevance, chosen by the designer.
series	ASCAAD
email
last changed	2021/08/09 13:13

_id	acadia21_502
id	acadia21_502
authors	Mytcul, Anna
year	2021
title	ARchitect
doi	https://doi.org/10.52842/conf.acadia.2021.502
source	ACADIA 2021: Realignments: Toward Critical Computation [Proceedings of the 41st Annual Conference of the Association of Computer Aided Design in Architecture (ACADIA) ISBN 979-8-986-08056-7]. Online and Global. 3-6 November 2021. edited by B. Bogosian, K. Dörfler, B. Farahi, J. Garcia del Castillo y López, J. Grant, V. Noel, S. Parascho, and J. Scott. 502-511.
summary	This research investigates gaming as a framework for design democratization in architecture, where the end user is the key decisionmaker in the design process. ARchitect is a multisensory game that promotes and explores the educational aspects of learning games and their influence on end user engagement with house co-design. This combinatorial game relies on an augmented reality (AR) application accessible through a smartphone, serving as a low-threshold tool for converting architectural drawings into 3D models in real time and using AR technology for design evaluation. By allowing for learning through playing, ARchitect provides alternative ways of gaining knowledge about design and architecture and empowers non-experts to take active and informed positions in shaping their future urban environments on a micro-scale, rethinking conventional market relations and exploring emerging personal and public values. The ARchitect game challenges conventional participatory design where an architect plays an essential role in facilitation of the design process and translation of end users’ design proposals. In contrast, the proposed game system allows non-architect players to autonomously produce and access design solutions through embedded computational simulation by an AR application, thus giving an equal chance to non-professionals to express their design visions and become aware of potential implications of their ideas. By providing free access to the game contents through the ARchitect platform and a playful user experience by which design principles can be learned, this game will inspire the general public to engage in conversation about home design, eventually spreading architectural literacy to less-privileged communities.
series	ACADIA
type	paper
email
last changed	2023/10/22 12:06

_id	sigradi2021_114
id	sigradi2021_114
authors	Cesar Rodrigues, Ricardo, Kenzo Imagawa, Marcelo, Rubio Koga, Renan and Bertola Duarte, Rovenir
year	2021
title	Big Data vs Smart Data on the Generation of Floor Plans with Deep Learning
source	Gomez, P and Braida, F (eds.), Designing Possibilities - Proceedings of the XXV International Conference of the Ibero-American Society of Digital Graphics (SIGraDi 2021), Online, 8 - 12 November 2021, pp. 217–228
summary	Due to the progressive growth of data dimensionality, addressing how much data and time is required to train deep learning models has become an important research topic. Thus, in this paper, we present a benchmark for generating floor plans with Conditional Generative Adversarial Networks in which we compare 10 trained models on a dataset of 80.000 samples, the models use different data dimensions and hyper-parameters on the training phase, beyond this objective, we also tested the capability of Convolutional Neural Networks (CNN) to reduce the dataset noise. The models' assessment was made on more than 6 million with the Frétche Inception Distance (FID). The results show that such models can rapidly achieve similar or even better FID results if trained with 800 images of 512x512 pixels, in comparison to high dimensional datasets of 256x256 pixels, however, using CNNs to enhance data consistency reproduced optimal results using around 27.000 images.
keywords	Floor plans, Generative design, Generative adversarial networks, Smart Data, Dataset reduction.
series	SIGraDi
email
last changed	2022/05/23 12:10

_id	ijac202119106
id	ijac202119106
authors	Del Campo, Matias; Alexandra Carlson, and Sandra Manninger
year	2021
title	Towards Hallucinating Machines - Designing with Computational Vision
source	International Journal of Architectural Computing 2021, Vol. 19 - no. 1, 88–103
summary	There are particular similarities in how machines learn about the nature of their environment, and how humans learn to process visual stimuli. Machine Learning (ML), more specifically Deep Neural network algorithms rely on expansive image databases and various training methods (supervised, unsupervised) to “make sense” out of the content of an image. Take for example how students of architecture learn to differentiate various architectural styles. Whether this be to differentiate between Gothic, Baroque or Modern Architecture, students are exposed to hundreds, or even thousands of images of the respective styles, while being trained by faculty to be able to differentiate between those styles. A reversal of the process, striving to produce imagery, instead of reading it and understanding its content, allows machine vision techniques to be utilized as a design methodology that profoundly interrogates aspects of agency and authorship in the presence of Artificial Intelligence in architecture design. This notion forms part of a larger conversation on the nature of human ingenuity operating within a posthuman design ecology. The inherent ability of Neural Networks to process large databases opens up the opportunity to sift through the enormous repositories of imagery generated by the architecture discipline through the ages in order to find novel and bespoke solutions to architectural problems. This article strives to demystify the romantic idea of individual artistic design choices in architecture by providing a glimpse under the hood of the inner workings of Neural Network processes, and thus the extent of their ability to inform architectural design.The approach takes cues from the language and methods employed by experts in Deep Learning such as Hallucinations, Dreaming, Style Transfer and Vision. The presented approach is the base for an in-depth exploration of its meaning as a cultural technique within the discipline. Culture in the extent of this article pertains to ideas such as the differentiation between symbolic and material cultures, in which symbols are defined as the common denominator of a specific group of people.1 The understanding and exchange of symbolic values is inherently connected to language and code, which ultimately form the ingrained texture of any form of coded environment, including the coded structure of Neural Networks.A first proof of concept project was devised by the authors in the form of the Robot Garden. What makes the Robot Garden a distinctively novel project is the motion from a purely two dimensional approach to designing with the aid of Neural Networks, to the exploration of 2D to 3D Neural Style Transfer methods in the design process.
keywords	Artificial intelligence, design agency, neural networks, machine learning, machine vision
series	journal
email
last changed	2021/06/03 23:29

_id	ecaade2021_254
id	ecaade2021_254
authors	Eisenstadt, Viktor, Arora, Hardik, Ziegler, Christoph, Bielski, Jessica, Langenhan, Christoph, Althoff, Klaus-Dieter and Dengel, Andreas
year	2021
title	Comparative Evaluation of Tensor-based Data Representations for Deep Learning Methods in Architecture
doi	https://doi.org/10.52842/conf.ecaade.2021.1.045
source	Stojakovic, V and Tepavcevic, B (eds.), Towards a new, configurable architecture - Proceedings of the 39th eCAADe Conference - Volume 1, University of Novi Sad, Novi Sad, Serbia, 8-10 September 2021, pp. 45-54
summary	This paper presents an extended evaluation of tensor-based representations of graph-based architectural room configurations. This experiment is a continuation of examination of recognition of semantic architectural features by contemporary standard deep learning methods. The main aim of this evaluation is to investigate how the deep learning models trained using the relation tensors as data representation means perform on data not available in the training dataset. Using a straightforward classification task, stepwise modifications of the original training dataset and manually created spatial configurations were fed into the models to measure their prediction quality. We hypothesized that the modifications that influence the class label will not decrease this quality, however, this was not confirmed and most likely the latent non-class defining features make up the class for the model. Under specific circumstances, the prediction quality still remained high for the winning relation tensor type.
keywords	Deep Learning; Spatial Configuration; Semantic Building Fingerprint
series	eCAADe
email
last changed	2022/06/07 07:55

_id	ijac202119205
id	ijac202119205
authors	Fukuda, Tomohiro; Marcos Novak, Hiroyuki Fujii, Yoann Pencreach
year	2021
title	Virtual reality rendering methods for training deep learning, analysing landscapes, and preventing virtual reality sickness
source	International Journal of Architectural Computing 2021, Vol. 19 - no. 2, 190–207
summary	Virtual reality (VR) has been proposed for various purposes such as design studies, presentation, simulation and communication in the field of computer-aided architectural design. This paper explores new roles for VR; in particular, we propose rendering methods that consist of post-processing rendering, segmentation rendering and shadow-casting rendering for more-versatile approaches in the use of data. We focus on the creation of a dataset of annotated images, composed of paired foreground-background and semantic-relevant images, in addition to traditional immersive rendering for training deep learning neural networks and analysing landscapes. We also develop a camera velocity rendering method using a customised segmentation rendering technique that calculates the linear and angular velocities of the virtual camera within the VR space at each frame and overlays a colour on the screen according to the velocity value. Using this velocity information, developers of VR applications can improve the animation path within the VR space and prevent VR sickness. We successfully applied the developed methods to urban design and a design project for a building complex. In conclusion, the proposed method was evaluated to be both feasible and effective.
keywords	Virtual reality, rendering, shader, deep learning, landscape analytics, virtual reality sickness, Fourth Industrial Revolution, computer-aided architectural design
series	journal
email
last changed	2024/04/17 14:29

_id	caadria2021_130
id	caadria2021_130
authors	Han, Yoojin and Lee, Hyunsoo
year	2021
title	Exploring the Key Attributes of Lifestyle Hotels: A Content Analysis of User-Created Content on Instagram
doi	https://doi.org/10.52842/conf.caadria.2021.1.071
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 71-80
summary	This study aims to investigate the key attributes of lifestyle hotels by analyzing user-created content on Instagram, an image-based social network service. In an era of uncertainty in the tourism and hospitality industry, it is inevitable that hotels must create a competitive identity. However, even with the significant growth of the lifestyle hotel segment, the concept of a lifestyle hotel is still vague. Therefore, to explore how to define, perceive, and interpret lifestyle hotels and to suggest their crucial attributes, this paper examines user-created content on Instagram. The data from 20,886 Instagram posts related to lifestyle hotels, including 2,209 locations, 43,586 hashtags, and 20,866 images, were analyzed using Vision AI, a social network analysis method and computer vision technology. The results of this study demonstrated that lifestyle hotels are perceived as design-focused branded hotels that represent the urban lifestyle and share both vacation and urban activities. Furthermore, the results reflected one of the latest hospitality trends-a holiday in an urban setting in addition to the primary purpose of traveling. Finally, this research suggests broader uses of big data and deep learning for analyzing how a place is consumed in a geospatial context.
keywords	Lifestyle Hotel; Hospitality Experiences; User-Created Content; Social Network Analysis; Vision AI
series	CAADRIA
email
last changed	2022/06/07 07:50

_id	caadria2021_117
id	caadria2021_117
authors	Ikeno, Kazunosuke, Fukuda, Tomohiro and Yabuki, Nobuyoshi
year	2021
title	Can a Generative Adversarial Network Remove Thin Clouds in Aerial Photographs? - Toward Improving the Accuracy of Generating Horizontal Building Mask Images for Deep Learning in Urban Planning and Design
doi	https://doi.org/10.52842/conf.caadria.2021.2.377
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 2, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 377-386
summary	Information extracted from aerial photographs is widely used in the fields of urban planning and architecture. An effective method for detecting buildings in aerial photographs is to use deep learning to understand the current state of a target region. However, the building mask images used to train the deep learning model must be manually generated in many cases. To overcome this challenge, a method has been proposed for automatically generating mask images by using textured 3D virtual models with aerial photographs. Some aerial photographs include thin clouds, which degrade image quality. In this research, the thin clouds in these aerial photographs are removed by using a generative adversarial network, which leads to improvements in training accuracy. Therefore, the objective of this research is to propose a method for automatically generating building mask images by using 3D virtual models with textured aerial photographs to enable the removable of thin clouds so that the image can be used for deep learning. A model trained on datasets generated by the proposed method was able to detect buildings in aerial photographs with an accuracy of IoU = 0.651.
keywords	Urban planning and design; Deep learning; Generative Adversarial Network (GAN); Semantic segmentation; Mask image
series	CAADRIA
email
last changed	2022/06/07 07:50

_id	sigradi2021_200
id	sigradi2021_200
authors	Karabagli, Kaan, Koc, Mustafa, Basu, Prithwish and As, Imdat
year	2021
title	A Machine Learning Approach to Translate Graph Representations into Conceptual Massing Models
source	Gomez, P and Braida, F (eds.), Designing Possibilities - Proceedings of the XXV International Conference of the Ibero-American Society of Digital Graphics (SIGraDi 2021), Online, 8 - 12 November 2021, pp. 191–202
summary	Machine learning (ML) has popular applications in domains involving image, video, text and voice. However, in architecture, image-based ML systems face challenges capturing the complexity of three-dimensional space. In this paper, we leverage a graph-based ML system that can capture the inherent topology of architectural conceptual designs and identify high-performing latent patterns within such designs. In particular, our goal is to translate architectural graph data into three-dimensional massing models. We are building on our prior ML work, where we, a. discovered latent topological features, b. composed building blocks into new designs, c. evaluated their feasibility, and d. explored Generative Adversarial (Neural) Networks (GAN)-generated design variations. We trained the ML system with architectural design data that we gathered from an online architectural design competition platform, translated them into machine-readable graph representations, and identified their essential subgraphs to develop novel compositions. In this paper, we explore how these novel designs (outputted in graph form), can be translated into three-dimensional architectural form. We present an ML approach to turn graph representations into functional volumetric massing models. The ultimate goal of the study is to develop an end-to-end pipeline to generate architectural design - from a graph representation to a fully developed conceptual proxy of a designed product. The research question is promising in automating conceptual design, and we believe the outcome can be relevant to other design disciplines as well.
keywords	Architectural design, machine learning, conceptual design, deep learning, artificial intelligence
series	SIGraDi
email
last changed	2022/05/23 12:10

_id	ecaade2021_037
id	ecaade2021_037
authors	Kikuchi, Takuya, Fukuda, Tomohiro and Yabuki, Nobuyoshi
year	2021
title	Automatic Diminished Reality-Based Virtual Demolition Method using Semantic Segmentation and Generative Adversarial Network for Landscape Assessment
doi	https://doi.org/10.52842/conf.ecaade.2021.2.529
source	Stojakovic, V and Tepavcevic, B (eds.), Towards a new, configurable architecture - Proceedings of the 39th eCAADe Conference - Volume 2, University of Novi Sad, Novi Sad, Serbia, 8-10 September 2021, pp. 529-538
summary	In redevelopment projects in mature cities, it is important to visualize the future landscape. Diminished reality (DR) based methods have been proposed to represent the future landscape after the structures are removed. However, two issues remain to be addressed in previous studies. (1) the user needs to prepare 3D models of the structure to be removed and the background structure to be rendered after removal as preprocessing, and (2) the user needs to specify the structure to be removed in advance. In this study, we propose a DR method that detects the objects to be removed using semantic segmentation and completes the removal area using generative adversarial networks. With this method, virtual removal can be performed without preparing 3D models in advance and without specifying the removal target in advance. A prototype system was used for verification, and it was confirmed that the method can represent the future landscape after removal and can run at an average speed of about 8.75 fps.
keywords	landscape visualization; virtual demolition; diminished reality (DR); deep learning; generative adversarial network (GAN); semantic segmentation
series	eCAADe
email
last changed	2022/06/07 07:52

_id	caadria2021_196
id	caadria2021_196
authors	Lu, Yueheng, Tian, Runjia, Li, Ao, Wang, Xiaoshi and Jose Luis, Garcia del Castillo Lopez
year	2021
title	CubiGraph5K - Organizational Graph Generation for Structured Architectural Floor Plan Dataset
doi	https://doi.org/10.52842/conf.caadria.2021.1.081
source	A. Globa, J. van Ameijde, A. Fingrut, N. Kim, T.T.S. Lo (eds.), PROJECTIONS - Proceedings of the 26th CAADRIA Conference - Volume 1, The Chinese University of Hong Kong and Online, Hong Kong, 29 March - 1 April 2021, pp. 81-90
summary	In this paper, a novel synthetic workflow is presented for procedural generation of room relation graphs of floor plans from structured architectural datasets. Different from classical floor plan generation models, which are based on strong heuristics or low-level pixel operations, our method relies on parsing vectorized floor plans to generate their intended organizational graph for further graph-based deep learning. This research work presents the schema for the organizational graphs, describes the generation algorithms, and analyzes its time/space complexity. As a demonstration, a new dataset called CubiGraph5K is presented. This dataset is a collection of graph representations generated by the proposed algorithms, using the floor plans in the popular CubiCasa5K dataset as inputs. The aim of this contribution is to provide a matching dataset that could be used to train neural networks on enhanced floor plan parsing, analysis and generation in future research.
keywords	Graph Theory; Algorithm; Architecture Design Dataset; Organizational Graph
series	CAADRIA
email
last changed	2022/06/07 07:59

_id	ecaade2021_148
id	ecaade2021_148
authors	Mintrone, Alessandro and Erioli, Alessio
year	2021
title	Training Spaces - Fostering machine sensibility for spatial assemblages through wave function collapse and reinforcement learning
doi	https://doi.org/10.52842/conf.ecaade.2021.1.017
source	Stojakovic, V and Tepavcevic, B (eds.), Towards a new, configurable architecture - Proceedings of the 39th eCAADe Conference - Volume 1, University of Novi Sad, Novi Sad, Serbia, 8-10 September 2021, pp. 17-26
summary	This research explores the integration of Deep Reinforcement Learning (RL) and a Wave Function Collapse (WFC) algorithm for a goal-driven, open-ended generation of architectural spaces. Our approach binds RL to a distributed network of decisions, unfolding through three key steps: the definition of a set of architectural components (tiles) and their connectivity rules, the selection of the tile placement location, which is determined by the WFC, and the choice of which tile to place, which is performed by RL. The act of thinking becomes granular and embedded in an iterative process, distributed among human and non-human cognitions, which constantly negotiate their agency and authorial status. Tools become active agents capable of developing their own sensibility while controlling specific spatial conditions. Establishing an interdependency with the human, that engenders the design patterns and becomes an indispensable prerequisite for the exploration of the generated design space, exceeding human or machinic reach alone.
keywords	Reinforcement Learning; Machine Learning; Proximal Policy Optimization; Assemblages; Wave Function Collapse
series	eCAADe
email
last changed	2022/06/07 07:58

For more results click below: