Pub Date : 2025-09-30DOI: 10.1016/j.foar.2025.08.008
Abhishek Mehrotra, Hwang Yi
Deep reinforcement learning (DRL) remains underexplored within architectural robotics, particularly in relation to self-learning of architectural design principles and design-aware robotic fabrication. To address this gap, we applied established DRL methods to enable robot arms to autonomously learn design rules in a pilot block wall assembly-design scenario. Recognizing the complexity inherent in such learning tasks, the problem was strategically decomposed into two sub-tasks: (i) target reaching (T1), modeled within a continuous action space, and (ii) sequential planning (T2), formulated within a discrete action space. For T1, we evaluated major DRL algorithms—Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), Deep Deterministic Policy Gradient, Twin Delayed Deep Deterministic Policy Gradient, and Soft Actor-Critic (SAC), and PPO, A2C, and Double Deep Q-Network (DDQN) were tested for T2. Performance was assessed based on training efficacy, reliability, and two novel metrics: degree index and variation index. Our results revealed that SAC was the best for T1, whereas DDQN excelled in T2. Notably, DDQN exhibited strong learning adaptability, yielding diverse final layouts in response to varying initial conditions.
{"title":"Performance comparison of deep reinforcement robot-arm learning in sequential fabrication of rule-based building design form","authors":"Abhishek Mehrotra, Hwang Yi","doi":"10.1016/j.foar.2025.08.008","DOIUrl":"10.1016/j.foar.2025.08.008","url":null,"abstract":"<div><div>Deep reinforcement learning (DRL) remains underexplored within architectural robotics, particularly in relation to self-learning of architectural design principles and design-aware robotic fabrication. To address this gap, we applied established DRL methods to enable robot arms to autonomously learn design rules in a pilot block wall assembly-design scenario. Recognizing the complexity inherent in such learning tasks, the problem was strategically decomposed into two sub-tasks: (i) target reaching (T1), modeled within a continuous action space, and (ii) sequential planning (T2), formulated within a discrete action space. For T1, we evaluated major DRL algorithms—Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), Deep Deterministic Policy Gradient, Twin Delayed Deep Deterministic Policy Gradient, and Soft Actor-Critic (SAC), and PPO, A2C, and Double Deep Q-Network (DDQN) were tested for T2. Performance was assessed based on training efficacy, reliability, and two novel metrics: degree index and variation index. Our results revealed that SAC was the best for T1, whereas DDQN excelled in T2. Notably, DDQN exhibited strong learning adaptability, yielding diverse final layouts in response to varying initial conditions.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1654-1680"},"PeriodicalIF":3.6,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499908","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-20DOI: 10.1016/j.foar.2025.08.010
Boyuan Yu , Jianing Luo , Kiwook Rha , Balsam Al-Hadithi , Zhiyong Li , Seohyeon Kim , Zijun Zhao , Yue Lu , Provides Ng
The nature–culture divide, a longstanding conceptual separation between human beings and the natural environment, is increasingly challenged by the pressing need to address climate change. This urgency calls for design approaches that can synthesise social and sustainable aspects, creating environmental-user-centric solutions. Our study aimed to bridge this divide by exploring the integration of digital and human crafts, with a focus on wood upcycling furniture as a case study. It investigates the flow of design information, creating an interactive feedback loop between physical and digital domains. To ensure the workflow aligns with stakeholder needs, the study engages professionals interdisciplinarily, including designers, informaticists, and engineers, to collectively test and reflect on the process. The proposed pipeline was then compared with the collaborative pipeline that emerged, incorporating stakeholder perspective to refine the system design. The resulting workflow embraced 3D scanning, AI-driven design generation, VR user scenario simulation, and AR-assisted physical fabrication. The digital and physical furniture prototypes suggest new avenues for design informatics by synthesising objective mathematical decisions with subjective semiotic inputs. By exploring the integration of human and machine crafts in the co-creation process, the reflections contribute to sustainable urban and community construction (SDG 11), revealing potentials for scalability in architectural production.
{"title":"Between social and sustainable: A collaborative wood upcycling design process integrating AI and mixed reality tools","authors":"Boyuan Yu , Jianing Luo , Kiwook Rha , Balsam Al-Hadithi , Zhiyong Li , Seohyeon Kim , Zijun Zhao , Yue Lu , Provides Ng","doi":"10.1016/j.foar.2025.08.010","DOIUrl":"10.1016/j.foar.2025.08.010","url":null,"abstract":"<div><div>The nature–culture divide, a longstanding conceptual separation between human beings and the natural environment, is increasingly challenged by the pressing need to address climate change. This urgency calls for design approaches that can synthesise social and sustainable aspects, creating environmental-user-centric solutions. Our study aimed to bridge this divide by exploring the integration of digital and human crafts, with a focus on wood upcycling furniture as a case study. It investigates the flow of design information, creating an interactive feedback loop between physical and digital domains. To ensure the workflow aligns with stakeholder needs, the study engages professionals interdisciplinarily, including designers, informaticists, and engineers, to collectively test and reflect on the process. The proposed pipeline was then compared with the collaborative pipeline that emerged, incorporating stakeholder perspective to refine the system design. The resulting workflow embraced 3D scanning, AI-driven design generation, VR user scenario simulation, and AR-assisted physical fabrication. The digital and physical furniture prototypes suggest new avenues for design informatics by synthesising objective mathematical decisions with subjective semiotic inputs. By exploring the integration of human and machine crafts in the co-creation process, the reflections contribute to sustainable urban and community construction (SDG 11), revealing potentials for scalability in architectural production.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1681-1696"},"PeriodicalIF":3.6,"publicationDate":"2025-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-05DOI: 10.1016/j.foar.2025.07.005
Jun Yin , Pengyu Zeng , Peilin Li , Jing Zhong , Tianze Hao , Han Zheng , Shuai Lu
In modern architectural design, as complexity increases and diverse demands emerge, reconstructing 3D spaces has become a crucial method. However, existing methods remain limited to small-scale scenarios and exhibit poor reconstruction accuracy when applied to building-scale environments, resulting in unstable mesh quality and reduced design productivity. Furthermore, the lack of real-time, interactive editing tools prolongs design iteration cycles and impedes workflow efficiency. To address this issue, we propose the following contributions:
(1) We construct ArchiNet++, an architectural dataset that includes 710,180 multi-view images, 5200 SketchUp models, and corresponding camera parameters from the conceptual design phase of architectural projects.
(2) We introduce Drag2Build++, an interactive 3D mesh reconstruction framework featuring drag-based editing and three core innovations: a differentiable geometry module for fine-grained deformation, a 2D-3D rendering bridge for supervision, and a GAN-based refinement module for photorealistic texture synthesis.
(3) Comprehensive experiments demonstrate that our model excels in generating high-quality 3D meshes and enables rapid mesh editing via drag-based interactions. Furthermore, by incorporating textured mesh generation into this interactive workflow, it improves both efficiency and modeling flexibility.
We hope this combination can contribute to a more intuitive modeling process and offer a practical tool set that supports the digital transformation efforts within architectural design.
{"title":"Drag2Build++: A drag-based 3D architectural mesh editing workflow based on differentiable surface modeling","authors":"Jun Yin , Pengyu Zeng , Peilin Li , Jing Zhong , Tianze Hao , Han Zheng , Shuai Lu","doi":"10.1016/j.foar.2025.07.005","DOIUrl":"10.1016/j.foar.2025.07.005","url":null,"abstract":"<div><div>In modern architectural design, as complexity increases and diverse demands emerge, reconstructing 3D spaces has become a crucial method. However, existing methods remain limited to small-scale scenarios and exhibit poor reconstruction accuracy when applied to building-scale environments, resulting in unstable mesh quality and reduced design productivity. Furthermore, the lack of real-time, interactive editing tools prolongs design iteration cycles and impedes workflow efficiency. To address this issue, we propose the following contributions:</div><div>(1) We construct <strong>ArchiNet++</strong>, an architectural dataset that includes 710,180 multi-view images, 5200 SketchUp models, and corresponding camera parameters from the conceptual design phase of architectural projects.</div><div>(2) We introduce <strong>Drag2Build++</strong>, an interactive 3D mesh reconstruction framework featuring drag-based editing and three core innovations: a differentiable geometry module for fine-grained deformation, a 2D-3D rendering bridge for supervision, and a GAN-based refinement module for photorealistic texture synthesis.</div><div>(3) Comprehensive experiments demonstrate that our model excels in generating high-quality 3D meshes and enables rapid mesh editing via drag-based interactions. Furthermore, by incorporating textured mesh generation into this interactive workflow, it improves both efficiency and modeling flexibility.</div><div>We hope this combination can contribute to a more intuitive modeling process and offer a practical tool set that supports the digital transformation efforts within architectural design.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1602-1620"},"PeriodicalIF":3.6,"publicationDate":"2025-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499905","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-30DOI: 10.1016/j.foar.2025.07.004
Yingle Zhang , Jiaqi He , Dhondub Dawa
This study investigates the architectural interventions of Francesco Venezia in the Belice Valley after the 1968 earthquake, with a particular focus on the c and the Open-Air Theatre in Salemi. Employing a qualitative, design-driven methodology, the research integrates formal spatial analysis with interpretative frameworks from spatial theory, cultural memory studies, and phenomenological approaches to architectural experience. Primary sources, including on-site surveys, original drawings, and project documentation, are complemented by critical essays and historical accounts. The analysis centers on the theme of spatio-temporal continuity, examining how Venezia's works engage with memory, ruins, and the fragmented identity of place. The findings reveal that Venezia's design process anchored in reinterpretation rather than reconstruction produces anomalous monuments that reestablish a sense of historical depth while resisting conventional forms of memorialization. His architecture articulates a dialectical relationship between absence and presence, solidifying a new spatial narrative in a landscape marked by trauma and displacement. This paper presents a globally applicable design paradigm for handling cultural memory, identity, and continuity in the architecture of crisis and recovery by suggesting a substitute for traditional post-disaster restoration.
{"title":"Spatio-temporal continuity in post-earthquake architecture of Francesco Venezia's anomalous monuments in the Belice Valley","authors":"Yingle Zhang , Jiaqi He , Dhondub Dawa","doi":"10.1016/j.foar.2025.07.004","DOIUrl":"10.1016/j.foar.2025.07.004","url":null,"abstract":"<div><div>This study investigates the architectural interventions of Francesco Venezia in the Belice Valley after the 1968 earthquake, with a particular focus on the c and the Open-Air Theatre in Salemi. Employing a qualitative, design-driven methodology, the research integrates formal spatial analysis with interpretative frameworks from spatial theory, cultural memory studies, and phenomenological approaches to architectural experience. Primary sources, including on-site surveys, original drawings, and project documentation, are complemented by critical essays and historical accounts. The analysis centers on the theme of spatio-temporal continuity, examining how Venezia's works engage with memory, ruins, and the fragmented identity of place. The findings reveal that Venezia's design process anchored in reinterpretation rather than reconstruction produces anomalous monuments that reestablish a sense of historical depth while resisting conventional forms of memorialization. His architecture articulates a dialectical relationship between absence and presence, solidifying a new spatial narrative in a landscape marked by trauma and displacement. This paper presents a globally applicable design paradigm for handling cultural memory, identity, and continuity in the architecture of crisis and recovery by suggesting a substitute for traditional post-disaster restoration.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1794-1809"},"PeriodicalIF":3.6,"publicationDate":"2025-08-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145500422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pre-sale and second-hand housing transaction modes dominate China's real estate market. However, many existing studies tend to treat the housing market as a homogeneous entity, overlooking the heterogeneity in core influencing factors across different transaction types. Thoroughly understanding the factors affecting various housing types can assist policymakers in formulating differentiated regulatory decisions through environmental intervention. Therefore, this study utilized multi-source big data and compared the performance of multiple machine learning models to evaluate the relative importance and nonlinear effects of building-level, neighborhood-level, and street-level built environment factors on pre-sale and second-hand housing prices. The empirical study of Chengdu, China revealed that distance to city center was the most significant explanatory factor influencing pre-sale and second-hand housing prices among all factors. Significant differences existed between neighborhood-level and street-level built environment factors' nonlinear and threshold effects on pre-sale and second-hand housing prices. Notably, subway accessibility showed a U-shaped impact on pre-sale housing prices. To the best of our knowledge, our study is one of the early studies systematically investigating the influencing differences between pre-sale housing prices and second-hand housing prices, providing robust evidence for regulating housing prices through environmental interventions and offering critical references for policymakers and market participants.
{"title":"Unraveling nonlinear relationship of built environment on pre-sale and second-hand housing prices using multi-source big data and machine learning","authors":"Qian Zeng , Hao Wu , Luyao Zhou , Xue Gao , Ningyuan Fei , Bart Julien Dewancker","doi":"10.1016/j.foar.2025.06.006","DOIUrl":"10.1016/j.foar.2025.06.006","url":null,"abstract":"<div><div>Pre-sale and second-hand housing transaction modes dominate China's real estate market. However, many existing studies tend to treat the housing market as a homogeneous entity, overlooking the heterogeneity in core influencing factors across different transaction types. Thoroughly understanding the factors affecting various housing types can assist policymakers in formulating differentiated regulatory decisions through environmental intervention. Therefore, this study utilized multi-source big data and compared the performance of multiple machine learning models to evaluate the relative importance and nonlinear effects of building-level, neighborhood-level, and street-level built environment factors on pre-sale and second-hand housing prices. The empirical study of Chengdu, China revealed that distance to city center was the most significant explanatory factor influencing pre-sale and second-hand housing prices among all factors. Significant differences existed between neighborhood-level and street-level built environment factors' nonlinear and threshold effects on pre-sale and second-hand housing prices. Notably, subway accessibility showed a U-shaped impact on pre-sale housing prices. To the best of our knowledge, our study is one of the early studies systematically investigating the influencing differences between pre-sale housing prices and second-hand housing prices, providing robust evidence for regulating housing prices through environmental interventions and offering critical references for policymakers and market participants.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1636-1653"},"PeriodicalIF":3.6,"publicationDate":"2025-07-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499907","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-16DOI: 10.1016/j.foar.2025.06.004
Xianchuan Meng , Jiadong Liang , Ximing Zhong
This paper presents a new generative artificial intelligence (AI) approach for creating modular skeletal frameworks, using vernacular bamboo stilt houses as examples to investigate an innovative methodological perspective. By transforming building skeletons to connected graphs, our method uses Variational Graph Autoencoders (VGAE) and Graph Sample and Aggregate (GraphSAGE) to generate 3D modular components based on spatial constraints set by users, such as axis grids and chosen room areas. The graph representation encodes structural elements as edges and their connections as nodes, maintaining critical dimensional constraints and spatial relationships. Using data from bamboo stilt houses built without architects, we make a specialized dataset of geometric skeletons for model training. Experimental results demonstrate the effectiveness of our approach in capturing the distribution of featured elements in building frameworks and in generating structurally sound designs, with GraphSAGE showing better performance compared to alternative methods. The probabilistic edge prediction approach allows for a collaborative human-AI design process, empowering designers while utilizing computational capabilities. The inherent flexibility of the graph-based representation makes it adaptable to a wide range of materials and scales.
{"title":"A generative artificial intelligence approach to modular skeletal framework modeling: Bamboo stilt houses as a case study","authors":"Xianchuan Meng , Jiadong Liang , Ximing Zhong","doi":"10.1016/j.foar.2025.06.004","DOIUrl":"10.1016/j.foar.2025.06.004","url":null,"abstract":"<div><div>This paper presents a new generative artificial intelligence (AI) approach for creating modular skeletal frameworks, using vernacular bamboo stilt houses as examples to investigate an innovative methodological perspective. By transforming building skeletons to connected graphs, our method uses Variational Graph Autoencoders (VGAE) and Graph Sample and Aggregate (GraphSAGE) to generate 3D modular components based on spatial constraints set by users, such as axis grids and chosen room areas. The graph representation encodes structural elements as edges and their connections as nodes, maintaining critical dimensional constraints and spatial relationships. Using data from bamboo stilt houses built without architects, we make a specialized dataset of geometric skeletons for model training. Experimental results demonstrate the effectiveness of our approach in capturing the distribution of featured elements in building frameworks and in generating structurally sound designs, with GraphSAGE showing better performance compared to alternative methods. The probabilistic edge prediction approach allows for a collaborative human-AI design process, empowering designers while utilizing computational capabilities. The inherent flexibility of the graph-based representation makes it adaptable to a wide range of materials and scales.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1621-1635"},"PeriodicalIF":3.6,"publicationDate":"2025-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499906","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-11DOI: 10.1016/j.foar.2025.06.002
Yichen Mo, Biao Li
As society confronts increasingly complex demands and the growing need for carbon-neutral architecture, AI-driven design methodologies are evolving rapidly. However, the lack of a unified integration platform in the design process continues to hinder AI's integration into real-world workflows. To address this challenge, we introduce ArchiWeb, a web-based platform specifically built to support AI-driven processes in early-stage architectural design. ArchiWeb transforms architectural representation and problem formulation by utilizing lightweight data protocols and a modular algorithmic network within an interactive web environment. Through its cloud-native, open-architecture framework, ArchiWeb enables deeper integration of AI technologies while accelerating the accumulation, sharing, and reuse of design knowledge across projects and disciplines. Ultimately, ArchiWeb aims to drive architectural design toward greater intelligence, efficiency, and sustainability—supporting the transition to data-informed, computationally enabled, and environmentally responsible design practices.
{"title":"ArchiWeb: A web platform for AI-driven early-stage architectural design","authors":"Yichen Mo, Biao Li","doi":"10.1016/j.foar.2025.06.002","DOIUrl":"10.1016/j.foar.2025.06.002","url":null,"abstract":"<div><div>As society confronts increasingly complex demands and the growing need for carbon-neutral architecture, AI-driven design methodologies are evolving rapidly. However, the lack of a unified integration platform in the design process continues to hinder AI's integration into real-world workflows. To address this challenge, we introduce ArchiWeb, a web-based platform specifically built to support AI-driven processes in early-stage architectural design. ArchiWeb transforms architectural representation and problem formulation by utilizing lightweight data protocols and a modular algorithmic network within an interactive web environment. Through its cloud-native, open-architecture framework, ArchiWeb enables deeper integration of AI technologies while accelerating the accumulation, sharing, and reuse of design knowledge across projects and disciplines. Ultimately, ArchiWeb aims to drive architectural design toward greater intelligence, efficiency, and sustainability—supporting the transition to data-informed, computationally enabled, and environmentally responsible design practices.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1551-1566"},"PeriodicalIF":3.6,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499902","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01DOI: 10.1016/j.foar.2025.05.006
Mikhael Johanes, Jeffrey Huang
The effectiveness of machine learning (ML) models for architectural applications relies on high-quality datasets balanced with advancements in model architecture and computational capacity. Current methods for evaluating architectural floor plan datasets typically depend on explicit semantic annotations, which limit their effectiveness and scalability when annotations are unavailable or inconsistent. To address this limitation, this research develops an isovist-based latent representation approach to quantitatively measure typicality and diversity within architectural datasets without relying on semantic labels. We introduce Isovist Latent Norm Typicality, a metric that leverages the statistical structure of latent representations derived from isovist morphological features using a variational autoencoder (VAE). This metric quantifies typicality by analyzing distributional shifts in latent representations between individual floor plans and a reference dataset using a modified Wasserstein distance. Experimental results demonstrate the approach's ability to distinguish typical from atypical floor plan configurations, capturing the morphological features that complement conventional metrics. Comparative analysis indicates that our method provides insights into spatial organization, correlating with conventional properties such as programmatic diversity and spatial openness. By quantifying typicality through purely morphological features, the proposed methodology facilitates dataset curation prior to costly semantic annotation, enhancing dataset quality and enabling scalability to more extensive and diverse architectural datasets.
{"title":"Latent distribution: Measuring floor plan typicality with isovist representation learning","authors":"Mikhael Johanes, Jeffrey Huang","doi":"10.1016/j.foar.2025.05.006","DOIUrl":"10.1016/j.foar.2025.05.006","url":null,"abstract":"<div><div>The effectiveness of machine learning (ML) models for architectural applications relies on high-quality datasets balanced with advancements in model architecture and computational capacity. Current methods for evaluating architectural floor plan datasets typically depend on explicit semantic annotations, which limit their effectiveness and scalability when annotations are unavailable or inconsistent. To address this limitation, this research develops an isovist-based latent representation approach to quantitatively measure typicality and diversity within architectural datasets without relying on semantic labels. We introduce <em>Isovist Latent Norm Typicality</em>, a metric that leverages the statistical structure of latent representations derived from isovist morphological features using a variational autoencoder (VAE). This metric quantifies typicality by analyzing distributional shifts in latent representations between individual floor plans and a reference dataset using a modified Wasserstein distance. Experimental results demonstrate the approach's ability to distinguish typical from atypical floor plan configurations, capturing the morphological features that complement conventional metrics. Comparative analysis indicates that our method provides insights into spatial organization, correlating with conventional properties such as programmatic diversity and spatial openness. By quantifying typicality through purely morphological features, the proposed methodology facilitates dataset curation prior to costly semantic annotation, enhancing dataset quality and enabling scalability to more extensive and diverse architectural datasets.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1585-1601"},"PeriodicalIF":3.6,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-31DOI: 10.1016/j.foar.2025.04.006
Yu Guo , Tianyu Fang , Zhe Cui , Rudi Stouffs
Architectural plan generation via pix2pix series algorithms faces dual challenges: the absence of domain-specific evaluation metrics and a lack of systematic insights into the joint impact of training configurations. To address the limitations of pix2pix-based models adaptation to architectural design, we designed a training regimen involving 12 experiments with varying training set sizes, dataset characteristics, and algorithms. These experiments utilized our self-built, high-quality, large-volume synthetic dataset of architectural-like plans. By saving intermediate models, we obtained 240 generative models for evaluation on a fixed test set. To quantify model performance, we developed a dual-aspect evaluation method that assesses predictions through pixel similarity (principle adherence) and segmentation line continuity (vectorization quality). Analysis revealed algorithm choice and training set size as primary factors, with larger sets enhancing the benefits of high-resolution and enhanced-annotation datasets. The optimal model achieved high-quality predictions, demonstrating strict adherence to predefined principles (0.81 similarity) and effective vectorization (0.86 segmentation line continuity). Testing on 7695 samples of varying complexity confirmed the model's robustness, strong generative capability, and controlled innovation within defined principles, validated through 3D model conversion. This work provides a domain-adapted framework for training and evaluating pix2pix-based architectural generators, bridging generative research and practical applications.
{"title":"A dual-aspect evaluation framework for architectural-like plan generation via pix2pix series algorithms","authors":"Yu Guo , Tianyu Fang , Zhe Cui , Rudi Stouffs","doi":"10.1016/j.foar.2025.04.006","DOIUrl":"10.1016/j.foar.2025.04.006","url":null,"abstract":"<div><div>Architectural plan generation via pix2pix series algorithms faces dual challenges: the absence of domain-specific evaluation metrics and a lack of systematic insights into the joint impact of training configurations. To address the limitations of pix2pix-based models adaptation to architectural design, we designed a training regimen involving 12 experiments with varying training set sizes, dataset characteristics, and algorithms. These experiments utilized our self-built, high-quality, large-volume synthetic dataset of architectural-like plans. By saving intermediate models, we obtained 240 generative models for evaluation on a fixed test set. To quantify model performance, we developed a dual-aspect evaluation method that assesses predictions through pixel similarity (principle adherence) and segmentation line continuity (vectorization quality). Analysis revealed algorithm choice and training set size as primary factors, with larger sets enhancing the benefits of high-resolution and enhanced-annotation datasets. The optimal model achieved high-quality predictions, demonstrating strict adherence to predefined principles (0.81 similarity) and effective vectorization (0.86 segmentation line continuity). Testing on 7695 samples of varying complexity confirmed the model's robustness, strong generative capability, and controlled innovation within defined principles, validated through 3D model conversion. This work provides a domain-adapted framework for training and evaluating pix2pix-based architectural generators, bridging generative research and practical applications.</div></div>","PeriodicalId":51662,"journal":{"name":"Frontiers of Architectural Research","volume":"14 6","pages":"Pages 1516-1535"},"PeriodicalIF":3.6,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145499900","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"艺术学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号