Abstract：Street space is the carrier of residents' public life. With the rapid development of big data, machine learning and other forms of artificial intelligence and multi-disciplinary linkage, the research of generating adversarial networks in the field of architecture and city has gradually become a new hotspot. Deep Generative Models in machine learning can learn to interpret data and generate designs autonomously. It provides a new perspective for improving the above problems. With this background, this study proposes a method to generate street style based on residents' preferences, and uses the “image-to-image translation” framework based on conditional GAN (cGAN) to connect 2D style pictures with the design of 3D architectural shapes, so as to establish a real-time feedback visualization platform for street transformation. When changing the height and form of the building model, the street style is generated autonomously according to the preferences of the residents. This study reveals the role of machine learning in street style generation, which can help users to design and evaluate urban street design schemes, and provide an important basis for the design of street space and architectural form.
王浩翼，杨钧然，吴子悦，张烨*. 机器学习视野下基于居民偏好的街道风格生成方法研究[J]. 新建筑, 2022, 40(6): 19-24.
WANG Haoyi,YANG Junran,WU Ziyue,ZHANG Ye. Street Style Generation Method Based on Residents' Preferences from the Perspective of Machine Learning. New Architecture, 2022, 40(6): 19-24.
版权所有 © 2017 新建筑杂志社