文章摘要
基于稳定扩散模型的皖南传统民居立面生成研究
Research on the Generation Method of Facade Images of Traditional Anhui South Residential Buildings Based on Stable Diffusion Model
投稿时间:2026-04-09  修订日期:2026-04-23
DOI:
中文关键词: 稳定扩散模型  LoRA  Frechet Inception Distance(简称FID)分数  因子分析  多元线性回归方程
英文关键词: Stable diffusion model  LoRA  Frechet Inception Distance(FID) score  factor analysis  multiple linear regression equation
基金项目:国家自然科学基金资助项目:居游共生视角下皖南非典型传统村落人因工效评估与空间优化应对研究(GrantNo. 52378001);安徽省高校协同创新项目:人工智能支持下城乡微更新设计生成技术开发与示范(GXXT-2023-047);安徽省优秀科研创新团队(Grant No.2022AH010021));安徽省联合共建学科重点实验室主任,编号:CSGX2024ZR01;安徽建筑大学引进人才和博士启动(项目编号:2023QDZ14),国家自然科学基金项目(面上项目,重点项目,重大项目)
作者单位邮编
李早 安徽建筑大学 建筑与规划学院 230601
徐梦薇 安徽建筑大学 建筑与规划学院 
汪强* 安徽建筑大学 建筑与规划学院 230601
蒋文倩 安徽建筑大学 建筑与规划学院 
摘要点击次数: 111
全文下载次数: 0
中文摘要:
      新时代乡村振兴背景下,皖南传统民居的保护与更新是当前研究的重要课题。梳理徽派建筑立面风貌要素特征作为稳定扩散模型的生成实验的基础,通过训练四个低秩适应(Low-Rank Adaptation, LoRA)模型进行传统村落立面生成实验,后对生成图像结果进行主客观评价,客观评价采用弗雷歇起始距离(Frechet Inception Distance, FID)分数对生成图像与真实图像对比判断生成样本的质量,主观评价则运用满意度评价的方法进行问卷调查,得到正向的结果反馈,运用因子分析方法对问卷结果数据归纳出四类影响因子,通过多元线性回归方程进一步探析。结果表明,形态美学因子对整体满意度影响最大。基于皖南立面要素特征的模型自动生成建筑立面,有助于突破传统设计力局限,探索一种高效且优质的村落民居建筑立面生成技术路径。主客观评价研究,为解决生成结果无法判断优劣提供了一种解决思路。
英文摘要:
      Under the background of rural revitalization in the new era, the preservation and renewal of traditional residential buildings in southern Anhui has become an important research topic. This study begins by sorting the fa?ade features of Huizhou-style architecture to serve as the foundation for generative experiments using a stable diffusion model. Four Low-Rank Adaptation (LoRA) models are trained to generate traditional village fa?ades. The generated image results are then evaluated through both subjective and objective methods. For objective evaluation, the Fréchet Inception Distance (FID) score is used to compare the generated images with real images, assessing the quality of the generated samples. For subjective evaluation, a satisfaction-based questionnaire survey is conducted, yielding positive feedback. Factor analysis is then applied to the survey data, identifying four categories of influencing factors, which are further analyzed using multiple linear regression. The results show that the morphological aesthetic factor has the greatest impact on overall satisfaction. The automatic generation of building fa?ades based on the characteristics of southern Anhui fa?ade elements helps break through the limitations of traditional design methods and explores an efficient, high-quality technical pathway for generating village residential fa?ades. The combined subjective and objective evaluation approach offers a solution to the challenge of assessing the quality of generated results.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭