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 4 of 4

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

_id	caadria2024_386
id	caadria2024_386
authors	Liang, Jiadong, Zhong, Ximing and Koh, Immanuel
year	2024
title	Bridging Bim and AI: A Graph-Bim Encoding Approach for Detailed 3D Layout Generation Using Variational Graph Autoencoder
doi	https://doi.org/10.52842/conf.caadria.2024.1.221
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. 221–230
summary	Building Information Modelling (BIM) data provides an abundant source with hierarchical and detailed information on architectural elements. Nevertheless, transforming BIM data into an understandable format for AI to learn and generate controllable and detailed three-dimensional (3D) models remains a significant research challenge. This paper explores an encoding approach for converting BIM data into graph-structured data for AI to learn 3D models, which we define as Graph-BIM encoding. We employ the graph reconstruction capabilities of a Variational Graph Autoencoder (VGAE) for the unsupervised learning of BIM data to identify a suitable encoding method. VGAE's graph generation capabilities also reason for spatial layouts. Results demonstrate that VGAE can reconstruct BIM 3D models with high accuracy, and can reason the entire spatial layout from partial layout information detailed with architectural components. The primary contribution of this research is to provide a novel encoding approach for bridging AI and BIM encoding. The Graph-BIM encoding method enables low-cost, self-supervised learning of diverse BIM data, capable of learning and understanding the complex relationships between architectural elements. Graph-BIM provides foundational encoding for training general-purpose AI models for 3D generation.
keywords	BIM, graph-structured, encoding approach, VGAE, graph reconstruction and generation
series	CAADRIA
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_378
id	caadria2024_378
authors	Zhong, Ximing, Liang, Jiadong, Pia, Fricker and Liu, Shengyu
year	2024
title	A Framework for Fine-Tuning Urban Gans Using Design Decision Data Generated by Architects Through Gans Applications
doi	https://doi.org/10.52842/conf.caadria.2024.2.019
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. 19–28
summary	Recent studies have utilized Generative Adversarial Networks (GANs) to learn from existing urban layouts for urban design tasks. We define these GANs as Urban-GAN. However, urban layouts generated by Urban-GAN lack specificity and often require multiple modifications by architects to meet specific design requirements, making the process inefficient and non-customizable. Inspired by the concept of fine-tuning language models, we propose a stacked GAN model framework that fine-tunes Urban-GAN using data generated by architects in solving specific design tasks, forming AD-Urban-GAN. Our results indicate that layouts produced by AD-Urban-GAN more effectively emulate architects' design morphology decisions, enhancing Urban-GAN’s adaptability and efficiency in handling design tasks. Furthermore, AD-Urban-GAN enhances the customizability of Urban-GAN models for specific urban design tasks, generating layouts that accurately understand and meet the requirements of specific tasks. AD-Urban-GAN significantly streamlines the process of generating design prototypes for specific task types, enabling precise quantitative control over urban layout results. This workflow establishes a data acquisition and training loop that strengthens the customizability of existing GANs. The design decision data generated by architects can improve the adaptability and customization of GANs models, facilitating efficient collaborative work between architects and artificial intelligence.
keywords	architect design decisions, Fine-tuning, GANs, Stack-GANs, adaptability, customizability
series	CAADRIA
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

No more hits.