Understanding architecture age and style through deep learning

Sun, Maoran; Zhang, Fan; Duarte, Fábio; Ratti, Carlo

doi:10.1016/j.cities.2022.103787

Cited by 52 publications

(26 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proposed model in this study, named FacMixNet, integrates a shared feature extraction architecture for the dual prediction of building age and built form two attributes that are conventionally classified separately. , The novel multitask learning framework posits the potential interrelation of features used for both age and built form recognition, as depicted in Figure Panel B.…”

Section: Methodsmentioning

confidence: 99%

Component-Level Residential Building Material Stock Characterization Using Computer Vision Techniques

Dai,

Jurczyk,

Arbabi

et al. 2024

Environ. Sci. Technol.

View full text Add to dashboard Cite

Residential building material stock constitutes a significant part of the built environment, providing crucial shelter and habitat services. The hypothesis concerning stock mass and composition has garnered considerable attention over the past decade. While previous research has mainly focused on the spatial analysis of building masses, it often neglected the component-level stock analysis or where heavy labor cost for onsite survey is required. This paper presents a novel approach for efficient component-level residential building stock accounting in the United Kingdom, utilizing drive-by street view images and building footprint data. We assessed four major construction materials: brick, stone, mortar, and glass. Compared to traditional approaches that utilize surveyed material intensity data, the developed method employs automatically extracted physical dimensions of building components incorporating predicted material types to calculate material mass. This not only improves efficiency but also enhances accuracy in managing the heterogeneity of building structures. The results revealed error rates of 5 and 22% for mortar and glass mass estimations and 8 and 7% for brick and stone mass estimations, with known wall types. These findings represent significant advancements in building material stock characterization and suggest that our approach has considerable potential for further research and practical applications. Especially, our method establishes a basis for evaluating the potential of component-level material reuse, serving the objectives of a circular economy.

show abstract

Section: Methodsmentioning

confidence: 99%

Component-Level Residential Building Material Stock Characterization Using Computer Vision Techniques

Dai,

Jurczyk,

Arbabi

et al. 2024

Environ. Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…In addition, comparing the accuracy with that achieved in the previous study demonstrated significant improvement over the results of using accurate Japanese real estate images with built-year and building-structure classifications of 0.367 and 0.786, respectively [17]. A previous study using SVI data achieved an accuracy of 0.869-0.871 in material classification in Chile [25] and an accuracy of 0.614 and 0.81 in built age classification in Austria and in Amsterdam, respectively [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30].…”

Section: B Classification Of the Built Year And Structure Of Individu...mentioning

confidence: 99%

“…When GIS building data and SVIs are combined for a highdensity area (such as an urban area), simply linking the nearest building and shooting point may not help to annotate the building accurately if the vehicle's travel direction and AOV are not considered. Previous studies used images recorded with the camera direction perpendicular to the travel direction, and this helped calculate the compass direction of each detected building easily based on its relative position in the image [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40]. However, a 360°SVI in a high-density building area cannot correctly annotate the buildings with GIS building data because the buildings are densely packed.…”

Section: A Automated Annotation For the Development Of Training Datamentioning

confidence: 99%

“…In recent years, Sun et al [29], [30] proposed a built-year estimation model using SVIs and presented an automated workflow for estimating building age from SVIs because of the increasing coverage of SVIs, such as GSV. The proposed model was constructed considering a 30-year time slice without considering the earthquake resilience of buildings; the low accuracy in dense areas within urban areas is due to the image cropping method.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Deep Learning Approach for Classifying the Built Year and Structure of Individual Buildings by Automatically Linking Street View Images and GIS Building Data

Ogawa

Zhao

Oki

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

“…For example, street view imagery prevails in understanding the urban physical construction environment and human habitat perception 22 , 23 . This data records abundant and various housing features from the pedestrian perspective, and are used to extract not only the physical objectives like housing height, density, and greening level 24 – 27 , but also indicators of subjective cognition of habitat comfortability, building style and street quality via computer vision and AI techniques 28 – 31 . All in all, street view imagery is viable for housing quality assessment of individual house automatically and at a large scale.…”

Section: Introductionmentioning

confidence: 99%

Combining deep learning and crowd-sourcing images to predict housing quality in rural China

Xu¹,

Gu²,

Chen³

et al. 2022

Sci Rep

View full text Add to dashboard Cite

Housing quality is essential to human well-being, security and health. Monitoring the housing quality is crucial for unveiling the socioeconomic development status and providing political proposals. However, depicting the nationwide housing quality in large-scale and fine detail is exceedingly rare in remote rural areas owing to the high cost of canonical survey methods. Taking rural China as an example, we collect massive rural house images for housing quality assessment by various volunteers and further build up a deep learning model based on the assessed images to realize an automatic prediction for huge raw house images. As a result, the model performance achieves a high R2 of 0.76. Afterward, the housing qualities of 10,000 Chinese villages are estimated based on 50,000 unlabeled geo-images, and an apparent spatial heterogeneity is discovered. Specifically, divided by Qinling Mountains-Huaihe River Line, housing quality in southern China is much better than in northern China. Our method provides high-resolution predictions of housing quality across the extensive rural area, which could be a complementary tool for automatical monitoring of housing change and supporting house-related policymaking.

show abstract

Understanding architecture age and style through deep learning

Cited by 52 publications

References 25 publications

Component-Level Residential Building Material Stock Characterization Using Computer Vision Techniques

Component-Level Residential Building Material Stock Characterization Using Computer Vision Techniques

Deep Learning Approach for Classifying the Built Year and Structure of Individual Buildings by Automatically Linking Street View Images and GIS Building Data

Combining deep learning and crowd-sourcing images to predict housing quality in rural China

Contact Info

Product

Resources

About