With Focus on Automated Labelling and Dataset Acquisition

Recently, Deep Learning (DL) has been widely used in the automation of urban layout processes. This study proposes a rule-based and Generative Adversarial Network (GAN) workflow to automatically select and label urban datasets to train customized GAN models for the generation of urban layout proposals. The developed workflow automatically collects and labels urban typology samples from the open-source map.  Furthermore, it controls the results of the GAN process through labeling and provides real-time urban layout suggestions in a co-design manner. The conducted case study shows that the average value of the GAN results, trained from an automatically generated dataset, meets the site's requirements. The developed co-design strategy allows the architect to control the GAN process to perform iterations of urban layout tasks. The research contributes to the found research gap in GAN applications in the field of urban design and planning. Many studies have demonstrated that training the GAN model by labeling enables machines to learn urban morphological features and urban layout logic.  However, two research gaps remain to be addressed in the previous studies. (1) The manual filtering of urban samples datasets for GAN meeting the site-specific design requirements is very time-consuming. (2) There is still a considerable challenge for the architecture to control the output of the GAN process to meet the design requirements without a suitable data labeling method.

Keywords: Deep Learning, Generative Adversarial Network (GAN), Urban Layout Process, Automatic Dataset Construction, Co-design

we publish our training models on this website: http://show.fujiazhiyu.cn/gantrans/modeltype

Limitations

The current atomized dataset labelling approach is currently only able to solve basic urban layout tasks. There is a need for enhanced methods to face complex urban layout tasks with GAN, such as the precise control of road widths, building boundaries, green areas etc. The combination of GAN methods and multi-objective optimization algorithms shows to be a possible solution. Also, more experiments and comparisons are needed to verify which urban morphology features can assist the architect in controlling the quality of urban layouts and which will create conflicts. Scale limitation is still a challenge for GAN application.  Although we applied two databases to generate layouts in different scales, it still cannot meet the requirements of multiple scales in the actual design process. In the current research, we testing for solving this problem with a rule-based algorithm combined with GAN.