Makine öğrenmesinin tasarım alanında kullanımı: grafik tabanlı GAN’lardan kütle modeli oluşturma

Öz

Bu makale, mimari grafik gösterimlerini üç boyutlu (3B) kavramsal kütle modellerine dönüştüren bir makine öğrenimi (ML) sistemi sunmaktadır. Öncelikle 2B kat planı oluşturmaya odaklanan önceki ML yaklaşımlarının aksine, yöntemimiz birden fazla bileşeni tek bir iş akışına entegre eder: (1) HouseGAN++ kullanarak grafik tabanlı girdi, (2) özel MATLAB görüntü işleme metotları ile görüntü tabanlı şekil çıkarma, (3) FloorplanToBlender ile 3B model oluşturma ve (4) görsel geliştirme için difüzyon model tabanlı stil transferi. Bu uçtan uca yaklaşım, otomatik plan-hacim dönüşümü ve üretken görüntü sentezi yoluyla estetik keşfin birleşiminde farklılık göstermektedir. Sonuçlar, geliştirilen sistemin manuel çabayı önemli ölçüde azaltırken verimli, çok aşamalı mimari fikir oluşturmayı sağladığını göstermektedir. Önerilen yöntem, kavram gelişimini hızlandırarak ve soyut mekânsal girdilerden stilistik olarak çeşitli çıktılar sunarak erken aşama tasarım süreçlerine katkıda bulunur.

Anahtar Kelimeler

• makine öğrenmesi
• kütle modelleri
• GAN
• konsept tasarım
• Grafik-tabanlı Derin Öğrenme

A Case Study of a Machine Learning Usage in Design: Conceptual Models from Graph-based GANs

Abstract

This paper presents a novel machine learning (ML) pipeline that transforms architectural graph representations into fully rendered three-dimensional (3D) conceptual massing models. Unlike previous ML approaches that focus primarily on 2D floorplan generation, our method integrates multiple components into a single workflow: (1) graph-based input using HouseGAN++, (2) image-based shape extraction via custom MATLAB processing, (3) 3D model construction with FloorplanToBlender, and (4) diffusion model–based style transfer for visual enhancement. This end-to-end approach is distinctive in its combination of automated plan-to-volume conversion and aesthetic exploration through generative image synthesis. The results show that our pipeline enables efficient, multi-stage architectural ideation while significantly reducing manual effort. The proposed method contributes to early-stage design processes by accelerating concept development and offering stylistically diverse outputs from abstract spatial inputs.

Keywords

• machine learning
• graph-based deep learning
• conceptual designs
• massing models
• GAN

References

1. As, İ., Pal, S., & Basu, P. (2018). Artificial Intelligence in Architecture: Generating Conceptual Design via Deep Learning. International Journal of Architectural Computing, 16(4), 306-327.
2. As, I., Pal, S., & Basu, P. (2019). Composing frankensteins: Data-driven design assemblies through graph-based deep neural networks. In the 107th Annual Meeting BLACK BOX: Articulating Architecture’s Core in the Post-Digital Era. Pittsburgh, PA, USA, ACSA
3. Atalay, M., Çelik, E. (2017). Artificial Intelligence and Machine Learning Applications in Big Data Analysis. Mehmet Akif Ersoy Üniversitesi Sosyal Bilimler Enstitüsü Dergisi 9 (22) 155-172.
4. Chaillou, S. (2019). AI + Architecture: Towards a New Approach. Harvard University, Graduate School of Design. Boston: Harvard University.
5. Egor, G., Sven, S., Martin, D., & Reinhard, K. (2019). Computer-aided approach to public buildings floor plan generation.Magnetizing Floor Plan Generator. 1st International Conference on Optimization-Driven Architectural Design (pp. 132-139). Amman: Elsevier Procedia.
6. freewayML. (2022). freewayML. Retrieved from freewayML: https://www.freewayml.com/about-us stability.ai. (2021). DreamStudio. Retrieved from DreamStudio: https://dreamstudio.ai
7. Gatsy, L. A., Ecker, A. S., & Bethge, M. (2015, September 2). arxiv. Retrieved from Cornell University: https://arxiv.org/abs/1508.06576
8. Grebtsew. (2021, May 7). FloorplanToBlender3d. Retrieved from Github: https://github.com/grebtsew/FloorplanToBlender3d

9. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Bengio, Y. (2014, June 10). Generative Adversarial Networks. Retrieved from arxiv: https://arxiv.org/abs/1406.2661
10. Harris, C. G., Stephens, M. J. (1988). A Combined Corner and Edge Detector. Alvey Vision Conference.
11. Karabağlı, K., Koç, M., Basu, P., As, İ. (2021). A Machine Learning Approach to Translate Graph Representations into Conceptual Massing Models. Sigradi 2021: Designing Possibilities, Ubiquitous Conference (pp. 191-202). Colombia: Perkins & Will
12. Nauata, N., Hosseini, S., Chang, K.-H., Chu, H., Cheng, C.-Y., & Furukawa, Y. (2021, March 3). House-GAN++: Generative Adversarial Layout Refinement Networks. Retrieved from Cornell University: https://arxiv.org/abs/2103.02574
13. OpenAI. (2021, January 5). DALL·E: Creating Images from Text. Retrieved from OpenAI: https://openai.com/blog/dall-e/
14. OpenCV. (2021, April 2). Image Segmentation with Watershed Algorithm. Retrieved from Open-Source Computer Vision: https://docs.opencv.org/4.5.2/d3/db4/tutorial_py_watershed.html
15. Pirouzaan. (2018, October 10). SYNTACTIC. Retrieved from food4rhino: https://www.food4rhino.com/en/app/syntactic
16. Rombach, R., Blattman, A., Lorenz, D., Esser, P., Ommer, B. (2022). High-Resolution Image Synthesis With Latent Diffusion Models. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
17. Senem, M. O., Koç, M., Tunçay, H. E., & As, İ. (2023). Using Deep Learning To Generate Front And Backyards In Landscape Architecture. Architecture and Planning Journal (APJ), 28(3), 1
18. Tamke, M., Nicholas, P., & Zwieryzycki, M. (2018). Machine Learning for architectural design: Practices and infrastructure. International Journal of Architectural Computing.
19. Topuz, B., Alp, N.Ç. (2023). Machine Learning in Architecture. Automation in Construction.
20. Tyagi, D. (2019, March 16). Introduction to Harris Corner Detector. Retrieved from Data Breach: https://medium.com/data-breach/introduction-to-harris-corner-detector-32a88850b3f6
21. Wang, Q. (2022). Machine Learning in Architecture. Lund: Lund University.
22. Wen, T., Tong, B., Liu, Y., Pan, T., Du, Y., Chen, Y., Zhang, S., (2022). Review of research on the instance segmentation of cell images. Computer Methods and Programs in Biomedicine.
23. Zhang, F., Nauata, N., & Furukawa, Y. (2020, March 31). Conv-MPN: Convolutional Message Passing Neural Network for Structured Outdoor Architecture Reconstruction. Cornell University: https://arxiv.org/abs/1912.01756.