Architectural Drawings Recognition and Generation through Machine Learning (original) (raw)
Related papers
Understanding and Visualizing Generative Adversarial Networks in Architectural Drawings
Learning, Prototyping and Adapting, Short Paper Proceedings of the 23rd International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 2018
Generative Adversarial Network (GAN) is a model framework in machine learning. It's specially used for learning and generating input and output data with similar or same format. PIX2PIXHD is a refined version of GAN, it's designed for learning image data in pairs, and generating predicted images based on the network model. The author applied PIX2PIXHD to learning and generating architectural drawings, marking rooms with different colours automatically by computer programs. Then, to understand how this network works, the author analysed the frame of the network, and give a detailed explanation about the three working principles of this network, convolution layer, residual network layer and deconvolution layer. Last, to visualize the network in architectural drawings, the author exported the data from each layer and each training epoch as grayscale images, finding that the features of architectural plan drawings have been learned step by step, and stored in the network as parameters. And the features in the drawings become more concise as the network goes deeper, and clearer as the training epoch increases. It might be inspiring comparing to the learning process of our human beings.
This study aims to produce Andrea Palladio's architectural plan schemes autonomously with generative adversarial networks(GAN), and to evaluate the plan drawing productions of GAN as a generative plan layout tool. GAN is a class of deep neural nets which is a generative model. In deep learning models, repetitive processes can be automated. Architectural drawing is a repetitive process in the act of architecture and plan drawing process can be made automated. For the automation of plan production system we used deep convolutional generative adversarial network (DCGAN) which is a subset of GAN models. And we evaluated the outputs of the DCGAN Palladian Plan scheme productions. Results show that not geometric similarities (shapes), but probabilistic models are at the centre of the generative system of GAN. For this reason, it should be kept in mind that while GAN algorithms are used as a generative system, they will produce statistically close visual models rather than geometrically close models. Nonetheless, GAN can generate both statistically and geometrically close models to the dataset. In first section we introduced a brief description about the place of the drawing in architecture field and future foresight of architecture drawings. In the second section, we gave detailed information about the literature on autonomous plan drawing systems. In the following sections, we explained the methodology of this study and the process of creating the plan drawing dataset, the algorithm training procedure, at the end we evaluated the generated plan schemes with rapid scene categorization and Frechet inception score.
Apartment Floor Plans Generation via Generative Adversarial Networks
Proceedings of the 25th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 2020
When drawing architectural plans, designers should always define every detail, so the images can contain enough information to support design. This process usually costs much time in the early design stage when the design boundary has not been finally determined. Thus the designers spend a lot of time working forward and backward drawing sketches for different site conditions. Meanwhile, Machine Learning, as a decision-making tool, has been widely used in many fields. Generative Adversarial Network (GAN) is a model frame in machine learning, specially designed to learn and generate image data. Therefore, this research aims to apply GAN in creating architectural plan drawings, helping designers automatically generate the predicted details of apartment floor plans with given boundaries. Through the machine learning of image pairs that show the boundary and the details of plan drawings, the learning program will build a model to learn the connections between two given images, and then the evaluation program will generate architectural drawings according to the inputted boundary images. This automatic design tool can help release the heavy load of architects in the early design stage, quickly providing a preview of design solutions for architectural plans.
2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021
Figure 1. The paper makes a breakthrough in the task of automated house layout generation. The right shows the mix of a ground-truth design made by an architect and our generated samples, based on the input bubble-diagram. Can you tell which one is the ground-truth? See the end of the caption for the answer. The paper proposes a novel generative adversarial layout refinement network, whose generator is trained to repeatedly apply and refine the design towards perfection. (The second sample from the right is the ground-truth.
International Congress on Human-Computer Interaction, Optimization and Robotic Applications”, 2023
We live in the age of Artificial Intelligence (AI) which permeates all aspects of our lives, from spam filtering to image classification on social media. While it is already wellestablished in industries ranging from heavy manufacturing to the IT field, its impact on the design professions remains relatively unexplored. This essay explores the use of neural networks in architecture, which is arguably the first genuinely 21st-century design technique and discusses experiments with Generative Adversarial Networks (GANs) to generate unexplored futuristic possible noble forms in architecture. In this way this paper also raises the question if machine can generate noble forms through its creative data optimization process. In this process one of the most famous heritages building of Bangladesh 60 dome mosque (Shat Gombuj Moshjid) has been examined to get expected result. Furthermore, this paper discusses how AI can be used as a personalized tool for architects to generate and express design ideas. It evaluates popular datasets for architectural purposes and considers the potential outcomes of experiments. The input of AI in the design process could usher in a new era of architectural design. As data continues to grow, it is shaping our collective future. Therefore, this paper concludes that it is essential to prepare our trained datasets to accept the future which might open up an extraordinary new chapter in the architectural realm.
GAN as a generative architectural plan layout tool: A case study for training DCGAN with Palladian Plans and evaluation of DCGAN outputs Üretken Mimari Plan Aracı Olarak GAN: Palladian Planları ile DCGAN Eğitimi ve DCGAN Çıktılarının Değerlendirilmesi için Bir Durum Çalışması, 2020
GAN as a generative architectural plan layout tool: A case study for training DCGAN with Palladian Plans and evaluation of DCGAN outputs REFERENCES Author(s) (2005). Ahmad, A. R., Basir, O. A., Hassanein, K., & Imam, M. H. (2004). Improved placement algorithm for layout optimization. In Proc. of the 2nd Int’l Industrial Engineering Conf.(IIEC’04). Boucher, B. (1998). Andrea Palladio: the architect in his time. Abbeville Press. Brock, A., Donahue, J., & Simonyan, K. (2018). Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096. Chaillou, S. (2019). AI & Architecture. Retrieved from https://towardsdatascience.com/ai-architecture-f9d78c6958e0 Dalgic, H. O., Bostanci, E., & Guzel, M. S. (2017). Genetic Algorithm Based Floor Planning System. arXiv preprint arXiv:1704.06016. Dinçer, A. E., Çağdaş, G., & Tong, H. (2014). Toplu Konutların Ön Tasarımı İçin Üretken Bir Bilgisayar Modeli. Megaron, 9(2). Donald, T. (1962). A Sumerian Plan In The John Rylands Library1. Journal of Semitic Studies, 7(2), 184-190. Duarte, J. P. (2005). A discursive grammar for customizing mass housing: the case of Siza's houses at Malagueira. Automation in Construction, 14(2), 265-275. Eastman, C. M. (1973). Automated space planning. Artificial intelligence, 4(1), 41-64. Foscari, A., Canal, B., & Façade, G. T. (2010). Andrea Palladio. Unbuilt Venice. Baden: Lars Muller Publishers. Generative adversarial network. (2019). Retrieved from https://en.wikipedia.org/wiki/Generative\_adversarial\_network Giaconi, G., Williams, K., & Palladio, A. (2003). The Villas of Palladio. Princeton Architectural. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. Grasl, T. (n.d.). GRAPE For Web - Shape grammar interpreter. Retrieved from http://grape.swap-zt.com/App/PalladianGrammar Grason, J. (1971, June). An approach to computerized space planning using graph theory. In Proceedings of the 8th Design automation workshop (pp. 170-178). ACM. Hemsoll, D. (2016). Palladian Design: The Good, the Bad and the Unexpected. Hillier, B., & Stonor, T. (2010). Space Syntax-Strategic Urban Design. City Planning Review, The City Planning Institute of Japan, 59(3), 285. Huang, W., & Zheng, H. (2018). Architectural drawings recognition and generation through machine learning. In Proceedings of the 38th Annual Conference of the Association for Computer Aided Design in Architecture, Mexico City, Mexico. Koning, H., & Eizenberg, J. (1981). The language of the prairie: Frank Lloyd Wright's prairie houses. Environment and planning B: planning and design, 8(3), 295-323. Krejcirik, M. (1969). Computer-aided plant layout. Computer-Aided Design, 2(1), 7-19. Levin, P. H. (1964). Use of graphs to decide the optimum layout of buildings. The Architects' Journal, 7, 809-815. Nagy, D., Lau, D., Locke, J., Stoddart, J., Villaggi, L., Wang, R., ... & Benjamin, D. (2017, May). Project Discover: An application of generative design for architectural space planning. In Proceedings of the Symposium on Simulation for Architecture and Urban Design (p. 7). Society for Computer Simulation International. Puppi, L. (1973). Andrea Palladio (Vol. 2). Milano: Electa. Puppi, L., Codato, P., Palladio, A., & Venchierutti, M. (2005). Andrea Palladio: introduzione alle architetture e al pensiero teorico. Arsenale. Radford, A., Metz, L., & Chintala, S. (2015). Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434. Ravenscroft, T. (2019). Wallgren Arkitekter and BOX Bygg create parametric tool that generates adaptive plans. Retrieved from https://www.dezeen.com/2019/06/27/adaptive-floor-plans-wallgren-arkitekter-box-bygg-parametric-tool/ Rojas, G. S., & Torres, J. F. (2006). Genetic algorithms for designing bank offices layouts. In Prosiding Third International Conference on Production Research–Americas’ Region. Rykwert, J., & Schezen, R. (1999). The palladian ideal. New York: Rizzoli. Stiny, G., & Mitchell, W. J. (1978). The palladian grammar. Environment and planning B: Planning and design, 5(1), 5-18. Weinzapfel, G., Johnson, T. E., & Perkins, J. (1971, June). IMAGE: an interactive computer system for multi-constrained spatial synthesis. In Proceedings of the 8th Design Automation Workshop (pp. 101-108). ACM. Wundram, M., Marton, P., & Pape, T. (1993). Andrea Palladio 1508-1580: Architect between the renaissance and baroque. Taschen,.
2022
In recent years, generative models have been rapidly transforming into a broad field of research, and artificial intelligence (AI) works are increasing. Since deep learning technologies such as Generative Adversarial Networks (GANs) providing synthesized new images are becoming more accessible, researchers in the design and related fields are very much interested in adapting GANs into practice. Especially, StyleGAN has a strong capability for image learning, reconstruction simulation, and absorbing the pixel characteristics of images in the input dataset. StyleGAN also produces similar imitation outputs and summarizes all the input data into one "average output". The study aims to reveal the potential of these outputs that can be employed as a visual inspiration aid for designers. This article will discuss the outputs of the experiments, findings, and prospects of StyleGAN.
Artificial Imagination Of Architecture With Deep Convolutional Neural Network
This paper attempts to determine if an Artificial Intelligence system using deep convolutional neural network (ConvNet) will be able to "imagine" architecture. Imagining architecture by means of algorithms can be affiliated to the research field of generative architecture. One of the many techniques of generating architecture is to break down the whole of a design into unitary elements and determine the rules that direct their associations into more complex compounded elements. This technique, called shape grammar, allows the system to generate accurate shapes with meaningful living spaces. The downside to this technique is the intellectual effort to specify all the grammar rules according to a set of unitary base elements. No doubt that imagining , in a top-down fashion, the whole tree of possibilities to classify all rules is a daunting task. Contrariwise, ConvNet makes it possible to avoid that difficulty by automatically extracting and classifying these rules as features from large example data. Moreover, image-base rendering algorithms can manipulate those abstract rules encoded in the ConvNet. From these rules and without constructing a prior 3D model, these algorithms can generate perspective of an architectural image. To conclude, establishing shape grammar with this automated system opens prospects for generative architecture with image-base rendering algorithms.
Perspectival GAN - Architectural form-making through dimensional transformation
Proceedings of the 40th International Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe) [Volume 1]
With the ascendance of Generative Adversarial Networks (GAN), promising prospects have arisen from the abilities of machines to learn and recognize patterns in 2D datasets and generate new results as an inspirational tool in architectural design. Insofar as the majority of ML experiments in architecture are conducted with imagery based on readily available 2D data, architects and designers are faced with the challenge of transforming machine-generated images into 3D. On the other hand, GAN-generated images are found to be able to learn the 3D information out of 2D perspectival images. To facilitate such transformation from 2D and 3D data in the framework of deep learning in architecture, this paper explores making new architectural forms from flat GAN images by employing traditional tools of projective geometry. The experiments draw on Brook Taylor's 19thcentury theorem of inverse projection system for creating architectural form from perspectival information learned from GAN images of Swiss alpine architecture. The research develops a parametric tool that automates the dimensional transformation of 2D images into 3D architectural forms. This research identifies potential synergic interactions between traditional tools and techniques of architects and deep learning algorithms to achieve collective intelligence in designing and representing creative architecture forms between humans and machines.
CAADRIA proceedings
This paper describes an approach of recognizing floor plans by assorting essential objects of the plan using deep-learning based style transfer algorithms. Previously, the recognition of floor plans in the design and remodeling phase was labor-intensive, requiring expert-dependent and manual interpretation. For a computer to take in the imaged architectural plan information, the symbols in the plan must be understood. However, the computer has difficulty in extracting information directly from the preexisting plans due to the different conditions of the plans. The goal is to change the preexisting plans to an integrated format to improve the readability by transferring their style into a comprehensible way using Conditional Generative Adversarial Networks (cGAN). About 100-floor plans were used for the dataset which was previously constructed by the Ministry of Land, Infrastructure, and Transport of Korea. The proposed approach has such two steps: (1) to define the important objects contained in the floor plan which needs to be extracted and (2) to use the defined objects as training input data for the cGAN style transfer model. In this paper, wall, door, and window objects were selected as the target for extraction. The preexisting floor plans would be segmented into each part, altered into a consistent format which would then contribute to automatically extracting information for further utilization.