基于GAN的地块建筑自动生成研究  

Automatic generation of plot buildings based on GAN

作　　者：吕新华 赖世娇 葛沿灵 范晨璟[2] 郝晟 Lyu Xinhua;Lai Shijiao;Ge Yanling;Fan Chenjing;Hao Sheng(Haian Natural Resources and Planning Bureau,Nantong 226600,China;Nanjing Forestry University,Nanjing 210037,China;Nanjing Yunyi Technology Co.,Ltd.,Nanjing 210019,China)

机构地区：[1]南通市海安市自然资源和规划局,江苏南通226600 [2]南京林业大学,江苏南京210037 [3]南京云亿科技有限公司,江苏南京210019

出　　处：《江苏科技信息》2023年第25期62-65,共4页Jiangsu Science and Technology Information

摘　　要：文章基于生成式对抗性神经网络(Generative Adversarial Nets,GAN)研究地块建筑自动生成,以AI辅助规划设计提高设计效率为目标,使用CycleGAN作为开发平台训练预处理模型,开发AI辅助制图软件介入城市规划设计任务。研究发现,生成式网络在建筑规划制图场景任务表现良好,可为未来建筑布局提供参考。该研究为城市规划及智慧城市建设提供了新视角和新思路。The article is based on Generative Adversarial Nets(GAN)to study the automatic generation of parcel buildings.With the goal of improving design efficiency through AI assisted planning and design,CycleGAN is used as a development platform to train preprocessing models,and AI assisted mapping software is developed to introduce urban planning and design tasks.Research has found that generative networks perform well in architectural planning and mapping scene tasks,providing reference for future architectural layout.This study provides new perspectives and ideas for urban planning and smart city construction.

关 键 词：GAN网络 建筑规划 CycleGAN 

分 类 号：TU984.14[建筑科学—城市规划与设计]