Scholarlink

<p>The early design phase of building projects has a critical impact on the construction phase. Recently, the generating and 3D reconstructing of single-storey interior layouts are progressed significantly by deep learning. However, the challenge of creating comprehensive 3D models of multi-storey residential buildings using multimodal data remains unexplored. This paper presents an innovative approach to automatically 3D building designs generation from using 2D layouts input and multimodal information. By leveraging the accessibility of 2D building data over 3D data. The approach uniquely integrates multimodal data with a Generative adversarial network (GAN), enhancing both the stability and originality of the generated outputs. Furthermore, a streamlined client interface is introduced that enables users to input specific design requirements and 3D models can be generated directly in Revit, thereby significantly improving user interaction and satisfaction. These advancements significantly enhance the precision and efficiency of early-stage building de-sign. It provides architects and designers with advanced tools for creating complex 3D models, optimizing spatial configurations, and improving design workflows.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

3D generative early-stage building design from 2D images: integration of multimodal data with GAN (3DMM-GAN)

  • 期刊论文

  • 发表日期：

    2025年4月6日

查看全文
  • Yaoli Xiong,
  • Changsaar Chai,
  • Yee Siang Gan,
  • Heap Yih Chong

摘要

The early design phase of building projects has a critical impact on the construction phase. Recently, the generating and 3D reconstructing of single-storey interior layouts are progressed significantly by deep learning. However, the challenge of creating comprehensive 3D models of multi-storey residential buildings using multimodal data remains unexplored. This paper presents an innovative approach to automatically 3D building designs generation from using 2D layouts input and multimodal information. By leveraging the accessibility of 2D building data over 3D data. The approach uniquely integrates multimodal data with a Generative adversarial network (GAN), enhancing both the stability and originality of the generated outputs. Furthermore, a streamlined client interface is introduced that enables users to input specific design requirements and 3D models can be generated directly in Revit, thereby significantly improving user interaction and satisfaction. These advancements significantly enhance the precision and efficiency of early-stage building de-sign. It provides architects and designers with advanced tools for creating complex 3D models, optimizing spatial configurations, and improving design workflows.