Scholarlink

<p>The traditional design of architectural floor plans is a manual and labor-intensive process prone to errors and inefficiencies. While digital tools such as CAD systems have streamlined certain aspects, the process remains heavily reliant on human expertise. In this study, we evaluate the use of Generative Adversarial Networks (GANs), specifically the pix2pix framework, for automating floor plan generation. We compare the performance of the pix2pix GAN model using Huber loss against traditional loss functions such as L1, L2, and Pinball loss. The Huber loss function, known for its robustness and resistance to outliers, demonstrates better stability during training and generates higher-quality floor plans. Additionally, we employ a conditional GAN to incorporate contextual information, improving the accuracy and realism of the generated floor plans. Experimental results show that the Huber loss-based model outperforms other loss functions in terms of reliability and efficiency for floor plan generation automation. This work represents a significant step toward automating the architectural design process and has the potential to revolutionize how floor plans are created and optimized. Code is available at <a href="https://github.com/ayushhsinghhh/Smart-Plan-AI">https://github.com/ayushhsinghhh/Smart-Plan-AI</a>.</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

Enhancing architectural floor plan generation with robust loss functions: a detailed evaluation of GAN-based approaches

  • 期刊论文

  • 发表日期：

    2025年5月27日

查看全文
  • Mohd Mohsin Ali,
  • Ayush Kumar Singh,
  • Manish Raj

摘要

The traditional design of architectural floor plans is a manual and labor-intensive process prone to errors and inefficiencies. While digital tools such as CAD systems have streamlined certain aspects, the process remains heavily reliant on human expertise. In this study, we evaluate the use of Generative Adversarial Networks (GANs), specifically the pix2pix framework, for automating floor plan generation. We compare the performance of the pix2pix GAN model using Huber loss against traditional loss functions such as L1, L2, and Pinball loss. The Huber loss function, known for its robustness and resistance to outliers, demonstrates better stability during training and generates higher-quality floor plans. Additionally, we employ a conditional GAN to incorporate contextual information, improving the accuracy and realism of the generated floor plans. Experimental results show that the Huber loss-based model outperforms other loss functions in terms of reliability and efficiency for floor plan generation automation. This work represents a significant step toward automating the architectural design process and has the potential to revolutionize how floor plans are created and optimized. Code is available at https://github.com/ayushhsinghhh/Smart-Plan-AI.