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 360

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

_id	caadria2024_68
id	caadria2024_68
authors	Chen, Changyu, Yu, Hanting and Guo, Yuhan
year	2024
title	Perception and Reality: Urban Green Space Analysis Using Language Model-Based Social Media Insights. A Case Study within Shanghai’s Inner Ring
doi	https://doi.org/10.52842/conf.caadria.2024.2.119
source	Nicole Gardner, Christiane M. Herr, Likai Wang, Hirano Toshiki, Sumbul Ahmad Khan (eds.), ACCELERATED DESIGN - Proceedings of the 29th CAADRIA Conference, Singapore, 20-26 April 2024, Volume 2, pp. 119–128
summary	As urbanization accelerates, urban green spaces play an increasingly integral role in daily life. The development of large language models (LLM) provides a practical method for assessing how people perceive green spaces. This study employs prompt-turning techniques to analyze social media data, aiming to uncover public sentiments and correlate them with the actual conditions of urban green spaces. By categorizing evaluative dimensions — location, landscape quality, facility levels, and management — diverse areas of public focus are revealed. We utilize both dictionary-based and language model-based methods for the analysis of overall emotional perception and dimensional emotional perception. Classification of sentiments into positive, neutral, and negative categories enhances our understanding of the public's general emotional inclinations. Using various spatial analysis techniques, the study delves into the current conditions of green spaces across these evaluative dimensions. In conclusion, a correlation analysis exposes patterns and disparities in these evaluative dimensions, providing valuable insights into understanding public emotional tendencies and offering effective recommendations from the perspective of public perception.
keywords	Urban green spaces, Social media analysis, Public Sentiment, Spatial analysis, Prompt-turning
series	CAADRIA
email
last changed	2024/11/17 22:05

_id	caadria2024_166
id	caadria2024_166
authors	Li, Jinmin, Luo, Yilu, Lu, Shuai, Zhang, Jingyun, Wang, Jun, Guo, Rizen and Wang, ShaoMing
year	2024
title	ChatDesign: Bootstrapping Generative Floor Plan Design With Pre-trained Large Language Models
doi	https://doi.org/10.52842/conf.caadria.2024.1.099
source	Nicole Gardner, Christiane M. Herr, Likai Wang, Hirano Toshiki, Sumbul Ahmad Khan (eds.), ACCELERATED DESIGN - Proceedings of the 29th CAADRIA Conference, Singapore, 20-26 April 2024, Volume 1, pp. 99–108
summary	Large language models (LLMs) have achieved remarkable success in various domains, revolutionizing tasks such as language translation, text generation, and question-answering. However, generating floor plan designs poses a unique challenge that demands the fulfilment of intricate spatial and relational constraints. In this paper, we propose ChatDesign, an innovative approach that leverages the power of pre-trained LLMs to generate floor plan designs from natural language descriptions, while incorporating iterative modifications based on user interaction. By processing user input text through a pre-trained LLM and utilizing a decoder, we can generate regression parameters and floor plans that are precisely tailored to satisfy the specific needs of the user. Our approach incorporates an iterative refinement process, optimizing the model output by considering the input text and previous results. Throughout these interactions, we employ many strategic techniques to ensure the generated design images align precisely with the user's requirements. The proposed approach is extensively evaluated through rigorous experiments, including user studies, demonstrating its feasibility and efficacy. The empirical results consistently demonstrate the superiority of our method over existing approaches, showcasing its ability to generate floor plans that rival those created by human designer. Our code will be available at https://github.com/THU-Kingmin/ChatDesign.
keywords	floor plan generation, large language models, user interactions, automatic design, deep learning, pre-train models
series	CAADRIA
email
last changed	2024/11/17 22:05

_id	ecaade2024_215
id	ecaade2024_215
authors	Park, Hyejin; Gu, Hyeongmo; Hong, Soonmin; Choo, Seungyeon
year	2024
title	Comparison of GAN-based Spatial Layout Generation Research Focusing on AIBIM-Spacemaker and GAN-based Prior Research
doi	https://doi.org/10.52842/conf.ecaade.2024.1.539
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 539–548
summary	Recent advancements in Large Language Models (LLM) and the emergence of ChatGPT are rapidly progressing Generative AI models, suggesting the possibility of AI replacing human creative activities. In architecture, where outcomes depend on human creative thinking, the pre-planning stage is crucial. Architectural planning involves decisions on mass, space layout, and space program, aiming for optimal design with a significant impact on subsequent stages. Creating a client-centric design within a given time prompts architects to search for diverse reference materials. However, finding comparable spatial layouts is challenging due to the predominant focus on materials, construction methods, and details. This study introduces AIBIM-Spacemaker, a Generative Adversarial Network (GAN)-based program we developed for generating spatial layouts through graphical composition of space programs. Focusing on a house with limited space usage but versatile layouts, the study collected 10,000 raster-based floor plan images, creating a training dataset annotated for spatial elements. Training this dataset using the YOLO model enabled automatic extraction of vector-based data representing spatial relationships from raster-based images. A GAN trained on this data resulted in AIBIM-Spacemaker, allowing users to create diverse spatial layouts. Executing a graph with nodes representing spaces and edges denoting relationships between doors and windows using the trained GAN produced varied spatial layouts. Verification, comparing actual ground truth values, GAN-generated outcomes, and architect-provided values confirmed the program's effectiveness in the planning stage. Performance was verified by comparing the program, learning method, dataset, and results developed in this study with previous studies on GAN-based spatial layout generation. This study identifies the potential for AI-based spatial layout generation, enhancing planning efficiency and contributing to intelligent design automation, with anticipated positive impacts on planning task efficiency.
keywords	Space Layout Generation, Space Program, Generative Adversarial Networks(GNN), You Only Look Once(YOLO), Pre-design stage
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_201
id	ecaade2024_201
authors	Hashizume, Keiji; Fukuda, Tomohiro; Yabuki, Nobuyoshi
year	2024
title	A Surface Modeling Method for Indoor Spaces from 3D Point Cloud Reconstructed by 3D Gaussian Splatting
doi	https://doi.org/10.52842/conf.ecaade.2024.1.695
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 695–704
summary	Building information modeling (BIM) is becoming increasingly important in architectural projects, and the implementation of BIM in new construction projects is progressing. On the other hand, many existing buildings do not have BIM data, so it is necessary to create it from scratch. A common method for converting existing buildings to BIM is scan-to-BIM, using techniques such as laser scanning or photogrammetry. However, laser scanning provides accurate point cloud data but requires expensive equipment, while photogrammetry is generally cost-effective but has lower accuracy point cloud data. Another approach for creating BIM from 2D images is to use neural radiance fields (NeRF). However, NeRF faces challenges in terms of data accuracy and processing speed when dealing with large or complex scenes. In contrast, 3D Gaussian Splatting is an emerging computer vision technology that uses machine learning to reconstruct 3D scenes from 2D images faster than NeRF, with comparable or better quality. Therefore, this study proposes a method to create surface models consisting of floors, walls, and ceilings as a preliminary step to creating BIM data for existing indoor spaces using 3D Gaussian Splatting. First, point cloud is generated using 3D Gaussian Splatting, followed by noise reduction. The point cloud is then classified based on height. Subsequently, processing such as extraction of boundary primitives from the point cloud of the floor and classification of feature points are performed to estimate the shape of the floor. Finally, ceilings and walls are created based on height and floor shape. The results of validation confirm an error of between 0.01m and 0.5m in the generated surface models. This study proposes a novel attempt to create 3D models using 3D Gaussian Splatting, contributing to the generation of BIM data for existing buildings.
keywords	Point Cloud, 3D Gaussian Splatting, Scan2BIM, Surface Modeling, Indoor 3D Reconstruction
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_252
id	ecaade2024_252
authors	Cui, Cassia De Lian; Cursi, Stefano; Simeone, Davide; Fioravanti, Antonio; Curra, Edoardo
year	2024
title	An Ontology-Driven Approach for Geometry Segmentation and Interpretation in Architectural Heritage/archaeology
doi	https://doi.org/10.52842/conf.ecaade.2024.2.239
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 2, pp. 239–248
summary	Digital information systems are widely used for heritage documentation and management activities. The creation of 3D models based on different survey techniques, like photogrammetry and laser scanning, allows a fast collection of the studied assets in the form of geometry dimensions and point clouds. However, the raw geometric information and the mesh/solid converted data need to be associated with semantic annotation, defined as external and formalized knowledge of the architectural artifact. This paper proposes a workflow using Semantic Web-related technologies to support point cloud segmentation activity of archaeological artifacts. The suggested approach is based on analyzing and integrating different layers of information through three main phases: the digital acquisition phase, the geometry creation phase, and the semantic enrichment phase. The defined framework is then applied to the archaeological case study of Tivoli to highlight how the workflow can significantly improve the quality and effectiveness of data segmentation in the existing heritage documentation processes by providing a solid basis for the generation of detailed and semantically enriched geometric information models. Finally, the creation of this system prototype will give overall support to aid the interpretations and value recognition of heritage sites thanks to the capability of representing and managing the categories (in Aristotle’s sense) and the uniqueness of concepts applied to this peculiar and paradigmatic case study.
keywords	Built heritage, Semantic annotation, Ontologies, Knowledge-based system
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_223
id	ecaade2024_223
authors	Kim, Taehoon; Kim, Geunjae; Hong, Soon Min; Choo, Seungyeon
year	2024
title	Development of Structure-Specific Architectural BIM Object Automatic Generation Technology for Reverse Design Based on Deep Learning
doi	https://doi.org/10.52842/conf.ecaade.2024.1.705
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 705–714
summary	This research developed a technology for classifying architectural objects based on point cloud data and creating Building Information Modeling (BIM) models in the reverse engineering process. This research analyzed the limitations in the process and current advancements in point cloud-based object recognition and classification technology, leveraging semantic segmentation. The classification method employed a semantic segmentation-based network to classify objects into desired classes within 3D point cloud data. Specifically, the TD3D network, known for its superior performance, was utilized in this study, with publicly available datasets used for training. Moreover, the developed algorithm for creating architectural object BIM models was specifically designed based on the simplest structure and form, namely reinforced concrete structure. In conclusion, the study aimed to develop technology more aligned with the fundamental purpose of performing reverse engineering in an architectural context. Analysis of validated architectural structures revealed that, despite deviating from actual measurement times, concrete-reinforced structures demonstrated the highest performance.
keywords	Reverse engineering, Deep Learning, Point Cloud, Automatic object generation, BIM
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_424
id	ecaade2024_424
authors	Yao, Chaowen; Fricker, Pia
year	2024
title	Neural Network-Driven 3D Generation of Urban Trees: Advancing carbon mitigation simulation through detailed tree modeling from point cloud data
doi	https://doi.org/10.52842/conf.ecaade.2024.1.605
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 605–614
summary	Urban digital twins are essential for climate-responsive urban planning but often fail to accurately represent trees, relying instead on oversimplified models that inadequately capture their environmental impact. Traditional methods for tree modeling, notably skeletonization, are both iterative and labor-intensive, leading to inefficiencies in environmental simulation accuracy. Addressing this gap, our study introduces a novel approach using a Block Sparse Convolutional Neural Network (BSCNN) to generate precise 3D tree models from mobile laser-scanned point clouds, significantly enhancing simulations for carbon mitigation efforts. Our method, tested in Helsinki's Jätkäsaari area, leverages pre-defined skeleton data to train the neural network, streamlining the extraction of movement direction and distance, thus bypassing traditional skeletonization's iterative nature. We further refine our model's accuracy and robustness by incorporating point clouds of varying densities and tailoring our approach to account for the morphological diversity of specific tree species. This specificity enables our models to more closely mirror real-world trees, making them invaluable for dynamic environmental modeling within urban digital twins. Moreover, our models support integration with the L-system, a prominent plant growth simulation algorithm, showcasing the potential of advanced neural networks to revolutionize computational architecture and foster precise, sustainable urban environmental simulations.
keywords	3D Point Cloud Analysis, Block Sparse Convolutional Neural Networks (BSCNN), Tree Morphology and Morphological Diversity, Urban Digital Twin and Environmental Simulation
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_117
id	ecaade2024_117
authors	Su, Xinyu; Luo, Jianhe; Liu, Zidong; Yan, Gaoliang
year	2024
title	Text to Terminal: A framework for generating airport terminal layout with large-scale language-image models
doi	https://doi.org/10.52842/conf.ecaade.2024.1.469
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 469–478
summary	Large-scale language-image (LLI) models present novel opportunities for architectural design by facilitating its multimodal process via text-image interactions. However, the inherent two-dimensionality of their outputs restricts their utility in architectural practice. Airport terminals, characterized by their flexibility and patterned forms, with most of the design operations occurring at the level of master plan, indicating a promising application area for LLI models. We propose a workflow that, in the early design phase, employs a fine-tuned Stable Diffusion model to generate terminal design solutions from textual descriptions and a site image, followed by a quantitative evaluation from an architectural expert's viewpoint. We created our dataset by collecting satellite images of 295 airport terminals worldwide and annotating them in terms of size and form. Using Terminal 2 of Zhengzhou Xinzheng International Airport as a case study, we scored the original and generated solutions on three airside evaluation metrics, verifying the validity of the proposed method. Our study bridges image generation and expert architectural design assessments, providing valuable insights into the practical application of LLI models in architectural practice and introducing a new method for the intelligent design of large-scale public buildings.
keywords	Multimodal Machine Learning, Diffusion Model, Text-to-Architecture, Airport Terminal Configuration Design, Post-Generation Evaluation
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_199
id	ecaade2024_199
authors	Zhong, Ximing; Liang, Jiadong; Li, Yingkai
year	2024
title	Building-Agent: A 3D generation agent framework integrating large language models and graph-based 3D generation model
doi	https://doi.org/10.52842/conf.ecaade.2024.2.291
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 2, pp. 291–300
summary	Large language models (LLMs) possess powerful intelligence, demonstrating unprecedented potential in AI-driven architectural design. While LLMs can understand design tasks, they lack the reasoning capability from language to three-dimensional (3D) architectural models. This paper proposes a novel 3D building generative agent framework, Building-Agent, which combines LLMs' decision-making capabilities with Graph Neural Networks (GNNs) generative abilities. Experiments utilize real design briefs and site constraints to test the building agent's task-processing capabilities. The results demonstrate that the Building-Agent can accurately predict different site layout outcomes and achieve high task completion rates. Furthermore, it enables interactive 3D building layout iteration through multi-step natural language instructions. The Building-Agent's ability to comprehend and reason about 3D spatial layouts, based on the graph representations of 3D models in the modeling engine and the requirements of natural language inputs, showcases its potential to accomplish tasks with initial proficiency. Compared to previous 3D generative models that rely on human decision-making for inputting spatial constraints, the Building-Agent paves the way for AI to comprehend and complete 3D design tasks autonomously, promising a transformative impact on AI and architectural design.
keywords	Building-Agent, Large Language Model, Graph Generation Model, Language Comprehending, 3D Spatial Reasoning, 3D Cognitive Ability
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_80
id	ecaade2024_80
authors	Li, Wenpei; Wu, Jiaqian; M. Herr, Christiane; Stouffs, Rudi
year	2024
title	Enhancing Lexicon Based Evaluation of Urban Green Space Characteristics and Perceptions with a Large Language Model
doi	https://doi.org/10.52842/conf.ecaade.2024.2.059
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 2, pp. 59–68
summary	Evaluating Urban green space Characteristics and Human Perceptions (UCHP) is crucial for landscape design and management due to their impact on public health. Online park reviews provide valuable insights into human-environment interactions, enabling the large-scale evaluation of UCHP. However, existing approaches to classify online park reviews commonly ignore text context, leading to low precision of UCHP quantification and supervised approaches are rarely applied due to huge cost. To improve the precision and effectiveness of UCHP quantification, we propose a novel workflow comprising five stages: custom lexicon creation, design of labels for a Large Language Model (LLM), sentence classification using lexicon and LLM, and performance evaluation using a manually annotated dataset and four metrics: precision, recall, accuracy, and F1 score. To examine the performance of the LLM, we compared the classification of 15 UCHP using LLM, lexicon, and lexicon+LLM. The analysis involved utilizing online park review sentences from Google Map and TripAdvisor using the proposed workflow. The higher precision, accuracy and F1 score demonstrate that combination of lexicon and LLM yields the highest performance, followed by using only lexicon and then solely LLM. This performance evaluation demonstrates the validity of the proposed LLM-aided workflow, providing a practical, reliable, and efficient alternative to the lower performance of unsupervised methods, or costly supervised classification methods. We discuss the limitations of lexicon+LLM and outline new opportunities for LLM application in landscape studies.
keywords	urban green space, characteristics and human perceptions, large language model, evaluation
series	eCAADe
email
last changed	2024/11/17 22:05

_id	architectural_intelligence2024_24
id	architectural_intelligence2024_24
authors	Giuseppe Bono
year	2024
title	Text-to-building: experiments with AI-generated 3D geometry for building design and structure generation
doi	https://doi.org/https://doi.org/10.1007/s44223-024-00060-5
source	Architectural Intelligence Journal
summary	The paper seeks to investigate novel potentials for building design and structure generation that arise at the intersection of computational design and AI-generated 3D geometries. Although the use of AI technologies is exponentially increasing inside the architectural discipline, the design of spatial building configurations using AI-generated 3D geometries is still limited in its applications and represents an ongoing field of investigation in advanced architectural research. In this regard, several questions still need to be answered: how can we design new building typologies from AI-generated 3D geometries? And how can we use these typologies to shape both the real and the virtual world? The paper proposes a new approach to architectural design where artificial intelligence is used as the starting point for design exploration, while computational design procedures are employed to convert AI-generated 3D geometries into building elements – such as columns, beams, horizontal and vertical surfaces. The paper starts with a general overview of the current use of artificial intelligence inside the architectural discipline, and then it moves towards the explanation of specific AI generative models for 3D geometry reconstruction and representation. Subsequently, the proposed working pipeline is analysed in more detail – from the creation of 3D geometries using generative AI models to the conversion of such geometries into building elements that can be further designed and optimised using computational design tools and methods. The results shown in the paper are achieved using Shap-E as the main AI model, though the proposed pipeline can be implemented with multiple AI models. The paper ends by showing some of the generated results, finally adding some considerations to the relationship between human and artificial creativity inside the architectural discipline. The work presented in the paper suggests that the use of computational design tools and methods combined with the tectonics of the latent space opens new opportunities for topological and typological explorations. In a time where traditional architectural typologies are moving towards stagnation due to their inability to satisfy new human needs and ways of living, exploring AI-based working pipelines related to architectural design allows the definition of new design solutions for the generation of new architectural spaces. In doing so, the serendipitous aspect of AI biases is used as an auxiliary force to inform design decisions, promoting the discovery of a new inbuilt dynamism between human and artificial creativity. In a time where AI is everywhere, understanding the measure of such dynamism represents a key aspect for the future of the architectural discipline.
series	Architectural Intelligence
email
last changed	2025/01/09 15:05

_id	ecaade2024_232
id	ecaade2024_232
authors	N. Panayiotou, Panayiotis; Kontovourkis, Odysseas
year	2024
title	A Holistic Documentation and Analysis of Timber Roof Structures in Heritage Buildings Using Scan to HBIM Approaches
doi	https://doi.org/10.52842/conf.ecaade.2024.1.715
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 715–724
summary	There is a great need for holistic documentation and management of heritage buildings using Historic Building Information Modelling (HBIM) frameworks. Limitations can be found in current literature regarding the accuracy, the level-of-detail, and the required attributes of final HBIM models, especially in cases where digital information intents to be used for the documentation of heritage timber roof structures. Previous research works indicate that geometry is created by the extrusion of the cross sections of the beams, and the usage of existing 2D drawings leading to simplified geometries in HBIM. This results in an absence of critical information, for example the bending of the wood, and its pathology. In this study, a novel Scan-to-HBIM methodology is exemplified and applied in heritage timber roof structures, which includes the implementation of recent remote sensing technologies for capturing the as-built data, with high levels of accuracy both in geometry as well as in pathology. In terms of geometry, algorithmic processes are used, that integrate parametric and BIM environments for the automatic creation of timber roof frames from point cloud data, which are adjustable to the abnormalities found in heritage buildings. As regards to pathology, high-resolution textured mesh models are created from photogrammetric procedures, which indicate in detail any possible defects to the existing timber elements. Detailed geometry and pathology are further analyzed, and a BIM database is created for documenting the typology, materiality, and level of damage to timber components. The methodology is tested on a Franko-Byzantine Timber roof Church in Cyprus, which includes a complex timber structural system.
keywords	Scan-to-HBIM, Terrestrial Laser Scanning, Photogrammetry, Algorithmic Design, Timber Roof Construction
series	eCAADe
email
last changed	2024/11/17 22:05

_id	caadria2024_414
id	caadria2024_414
authors	Uzun, Fatih, Altun, Sevgi, Kayasü, Sena, Öztürk, Berkay, ªahin, Yusuf, Ünal, Gözde and Özkar, Mine
year	2024
title	Utilizing UV-Mapping for the 3D-Point Cloud Segmentation of Architectural Heritage
doi	https://doi.org/10.52842/conf.caadria.2024.2.313
source	Nicole Gardner, Christiane M. Herr, Likai Wang, Hirano Toshiki, Sumbul Ahmad Khan (eds.), ACCELERATED DESIGN - Proceedings of the 29th CAADRIA Conference, Singapore, 20-26 April 2024, Volume 2, pp. 313–322
summary	This paper presents a procedural framework to process raw photogrammetric data intended for 3D point cloud segmentation of masonry structures. Raw data is segmented in order to obtain operable 3D models through processes that increasingly integrate computational methods, which can improve efficiency and accuracy. The idea is to improve the quality of the initial data, which can then be used to train a machine-learning system in identifying these materials more accurately. The approach incorporates a high-poly detailed mesh model generated through photogrammetry. The detailed model serves as a reference to extract colour information that we project onto a custom-created, low-poly representation of the dome architectural element, ensuring a precise fit with the target model. The model to utilise UV maps and height maps to preprocess data across various scales is a step towards facilitating the documentation and conservation of historic structures with an awareness of architectural knowledge.
keywords	architectural heritage, unit-based, masonry documentation, 3D point cloud segmentation, UV mapping, point cloud processing
series	CAADRIA
email
last changed	2024/11/17 22:05

_id	ecaade2024_253
id	ecaade2024_253
authors	Ke, Daijia; Agniputra, Akbar; Feng, Zhaoyan; Wu, Ilin; Di Carlo, Ilaria; Papeschi, Annarita
year	2024
title	Cartographies of Immersive Fractality: An exploration of collective emotive responses in urban settings through Machine Learning
doi	https://doi.org/10.52842/conf.ecaade.2024.1.519
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 519–528
summary	Recent advances in machine learning technologies offer avenues for a more efficient analysis of large photographic and text-based datasets, facilitating a deeper understanding of the fundamental characteristics inherent in the immersive representation of the urban environment. It is known that automatic fractal processing in the human visual system triggers positive emotive responses to the environment. The project explores the correlation among fractal aesthetics, visual perception, and emotional responses in urban settings, developing an integrated evaluation method that uses the data-scraping of existing online photographic media from Flickr and Google Street View (GSV). Taking the area of Southbank in London (UK) as a case study, the study initially employed a sentiment analysis method rooted in the Lexical dictionary from TextBlob. Further, an extensive online GSV urban scenery dataset was built via Google API. The photographic dataset was then evaluated by fractal dimension as a quantitative index to measure the complexity of fractal patterns. Concurrently, to enhance the comprehension of the composition of urban form, a semantic segmentation method for image analysis was implemented. A comparative evaluation of the data collected indicated the key role of fractal patterns described by vegetation in the generation of positive emotional responses, underscoring with methodological rigour the potentially transformative impact of the experience of fractal patterns and green infrastructures in open urban spaces.
keywords	Visual perception, Sentiment analysis, Psychogeography, Fractal aesthetics, Machine Learning
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_198
id	ecaade2024_198
authors	Liang, Jiadong; Zhong, Ximing; Koh, Immanuel
year	2024
title	Building-VGAE: Generating 3D detailing and layered building models from simple geometry
doi	https://doi.org/10.52842/conf.ecaade.2024.1.625
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 625–634
summary	In the current field of AI-assisted architectural design, deep learning models primarily focus on simulating the highly detailed final models designed by human architects. However, in practical design tasks, the final model demands a high level of detail and clear layered classification information for building components. This presents a more significant challenge. We propose a three-dimensional(3D) building generation framework—Building-VGAE, based on Variational Graph Autoencoder (VGAE). Building-VGAE can generate 3D models with detailed building components and layered structure information from end to end, according to design constraints and building volumes. Building-VGAE’s experiment involves transforming 27,965 Housegan data into 3D data represented as graph-structured. The VGAE model then learns the data features and predicts the building component categories to which nodes and edges belong in the experiment. The results demonstrate that the framework can precisely reconstruct and predict building layouts that comply with design constraints and enable unified editing of building components of the same category. Building-VGAE contributes to its ability to learn the generative relationship from design constraints and building volumes to complex high-detail models compared to existing AI generative models. It also possesses prediction and editing capabilities based on the layered classification information of building components. This framework has the potential to position AI as a design partner for human architects, offering end-to-end 3D generative intelligence.
keywords	Variational Graph Auto-Encoder, 3D Spatial Grid Structure, Detailed Building Components, Layered Structure, Graph Reconstruction and Generation
series	eCAADe
email
last changed	2024/11/17 22:05

_id	caadria2024_531
id	caadria2024_531
authors	Mottaghi, Esmaeil, Abuzuraiq, Ahmed M. and Erhan, Halil
year	2024
title	D-Predict: Integrating Generative Design and Surrogate Modelling with Design Analytics
doi	https://doi.org/10.52842/conf.caadria.2024.1.455
source	Nicole Gardner, Christiane M. Herr, Likai Wang, Hirano Toshiki, Sumbul Ahmad Khan (eds.), ACCELERATED DESIGN - Proceedings of the 29th CAADRIA Conference, Singapore, 20-26 April 2024, Volume 1, pp. 455–464
summary	The increasing importance of performance prediction in architecture has driven designers to incorporate computational tools like generative design and building simulations to widen and guide their exploration. However, these tools pose their own challenges; specifically, simulations can be computationally demanding and generative design leads to large design spaces that are hard to navigate. To address those challenges, this paper explores integrating machine learning-based surrogate modelling, interactive data visualisations, and generative design. D-Predict, a prototype, features the generation, management and comparison of design alternatives aided with surrogate models of daylighting and energy.
keywords	generative design, building performance assessment, surrogate modelling, machine learning, design analytics
series	CAADRIA
email
last changed	2024/11/17 22:05

_id	caadria2024_293
id	caadria2024_293
authors	Xu, Weishun, Li, Mingming and Yang, Xuyou
year	2024
title	Can Generative AI Models Count? Finetuning Stable Diffusion for Architecture Image Generation with Designated Floor Numbers Using a Small Dataset
doi	https://doi.org/10.52842/conf.caadria.2024.1.089
source	Nicole Gardner, Christiane M. Herr, Likai Wang, Hirano Toshiki, Sumbul Ahmad Khan (eds.), ACCELERATED DESIGN - Proceedings of the 29th CAADRIA Conference, Singapore, 20-26 April 2024, Volume 1, pp. 89–98
summary	Despite the increasing popularity of off-the-shelf text-to-image generative artificial intelligence models in early-stage architectural design practices, general-purpose models are challenged in domain-specific tasks such as generating buildings with the correct number of floors. We hypothesise that this problem is mainly caused by the lack of floor number information in standard training sets. To overcome the often-dodged problem in creating a text-image pair dataset large enough for finetuning the original model in design research, we propose to use BLIP method for both understanding and generation based automated labelling and captioning with online images. A small dataset of 25,172 text-image pairs created with this method is used to finetune an off-the-shelf Stable Diffusion model for 10 epochs with affordable computing power. Compared to the base model with a less than 20% chance to generate the correct number of floors, the finetuned model has an over 50% overall chance for correct floor number and 87.3% change to control the floor count discrepancy within 1 storey.
keywords	text-to-image generation, model finetuning, stable diffusion, automated labelling
series	CAADRIA
email
last changed	2024/11/17 22:05

_id	ecaade2024_242
id	ecaade2024_242
authors	Koh, Immanuel; Saw, Man Lin
year	2024
title	Architectural Dramaturgy: A total and endless theatre with multimodal artificial intelligence
doi	https://doi.org/10.52842/conf.ecaade.2024.1.559
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 559–564
summary	Close to a century ago, Walter Gropius proposed the highly influential but unrealised project ‘Total Theatre’ (1927). The project captured Gropius’ adapted formulation of ‘Gesamtkunstwerk’(‘total work of art’) as ‘Total Design’ within the Bauhaus, and more specifically, in relation to ‘The Theatre of the Bauhaus’ (1924). The ‘Total Theatre’ was Gropius’ attempt to dynamically reconfigure the previously static relationship between actors and audience into one that is participatory and filmic in spatial layout. Around the same time in history, shortly after designing the set for Eugene O’Neill’s ‘The Emperor Jones’ in 1924, Frederick Kiesler began to conceptualise his ‘Endless Theatre’ (1926). In the former, Kiesler created a constructivist stage space with moving walls/floors/ceilings animating in-sync with the play’s narrative and producing cinematographic effects similar to those found in films. For both Gropius and Kiesler, the theatre is a fertile ground for incorporating the latest technologies to prototype a ‘totally’ multimodal and ‘endlessly’ generative form of architecture. Against the backdrop of such past intellectual efforts in architecture, and that of today’s rapidly foregrounding of Large Language Models (LLMs) and Large Multimodal Models (LMMs) in artificial intelligence (AI) within the architecture discipline, the paper posits that it is timely to revisit this conceptual cross-fertilisation of theatre and architecture – architectural dramaturgy. In doing so, the research aims to extend Gropius’s and Kiesler’s concept of the ‘total’ and the ‘endless’ respectively through the lens of the ‘computational’. The first experiment uses different datasets from existing established theatre creators to train and prompt-engineer relevant AI models, namely a text-to-text model for playwright Arthur Miller, a text-to-image model for director Ivo van Hove, and a text-to-audio model for composer Stephen Sondheim. In improving the coherence of the results, the second experiment replaces unimodality with multimodality leveraging a single source of video data (the poignant “To be or not to be” soliloquy from Shakespeare’s Hamlet) to formulate a human-in-the-loop interpretative framework by utilising a combination of text-image-to-image model and text-image-to-video model, and further postprocessed with image inpainting model and image-to-3D model. With the deliberate bracketing of site, programmes and other specificities typical of an architecture project, the research demonstrates how concepts borrowed from theatre when layered with multimodal AI could extend the discipline’s longstanding conception of a total and endless architecture.
keywords	Theatre, Deep Learning, Large Multimodal Models, Kiesler, Gropius
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ecaade2024_297
id	ecaade2024_297
authors	Massafra, Angelo; Coraglia, Ugo Maria; Predari, Giorgia; Gulli, Riccardo
year	2024
title	Building Information Model Analysis Through Large Language Models and Knowledge Graphs
doi	https://doi.org/10.52842/conf.ecaade.2024.1.685
source	Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 685–694
summary	The advent of Large Language Models (LLMs) seems to mark a break between past and present in the methods of structuring knowledge, making it possible today to transfer this capability to machines even in a sector like AECO, always been information-intensive but resistant to technological transition. In terms of knowledge, the most established paradigm has been Building Information Modelling (BIM), with IFC functioning as the main schema for standardizing the industry's information. Added to this are knowledge graphs that, emerging with semantic web technologies, allow storing knowledge in structures consisting of nodes and edges with semantic meanings. Nevertheless, a barrier to the widespread adoption of BIM is its accessibility. Querying BIM models is often limited for stakeholders without digital skills, who may struggle to access the vast amount of information stored in these complex informative models. In an attempt to outline one of the possible uses of LLMs in BIM, this research proposes a method for querying BIM models through textual prompts aimed at analyzing a selected case study. In the workflow, a BIM model is first realized. Then, data is integrated into a knowledge graph. Next, ChatGPT's LLMs are used to activate query functions for the analysis of the graph. The results of the queries are displayed in a user-friendly graphical user interface. The study's outcomes offer insights for researchers and industry professionals, highlighting emerging research potentials for LLMs in the field.
keywords	Building Information Modeling, Large Language Models, Natural Language Processing, Knowledge Graphs
series	eCAADe
email
last changed	2024/11/17 22:05

_id	ijac202322203
id	ijac202322203
authors	Yang, Stephen; Jonathan Dortheimer, Aaron Sprecher and Qian Yang
year	2024
title	When design workshops meet chatbots: Meaningful participation at scale?
source	International Journal of Architectural Computing 2024, Vol. 22 - no. 2, 1-22
summary	This paper explores the potential of chatbots, powered by large language models, as a tool for fostering community participation in architectural and urban design. By taking a hybrid approach to community participation in a real-world mixed-use building project, in which we integrated remote chatbot engagements with face-to-face workshops, we explored the potential for a hybrid approach to scaling up the reach of participation while ensuring that such participation is meaningful, genuine, and empowering. Our findings suggest that a hybrid approach amplified the strengths and mitigated the shortcomings of the two methods. The chatbot was effective in sustaining the length of participation, broadening the reach of participation, and creating a personalized environment for introspection. Meanwhile, the face-to-face workshops still played a crucial role in bolstering community ties and trust. This research contributes to understanding chatbots’ strengths and weaknesses in participatory processes, both within spatial design and beyond. In addition, it informs future explorations of participatory processes that span different spatial-temporal configurations
keywords	Artificial intelligence, chatbot, community participation, large language models, natural language processing, participatory design
series	journal
last changed	2024/07/18 13:03

For more results click below: