| 李早,徐梦薇,汪强,蒋文倩.基于稳定扩散模型的皖南传统民居立面生成研究[J].安徽建筑大学学报,2026,34(3):1-8 |
| 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 onStable Diffusion Model |
| |
| DOI： |
| 中文关键词: 皖南传统民居 LoRA FID分数 因子分析 多元线性回归方程 |
| 英文关键词: traditional dwellings of southern Anhui LoRA Fréchet Inception Distance（FID）score factor analysis multiple linearregression equation |
| 基金项目:国家自然科学基金项目（52378001）；安徽省高校协同创新项目（GXXT-2023-047）；安徽省优秀科研创新团队项目（2022AH010021）；安徽省联合共建学科重点实验室主任基金项目（CSGX2024ZR01）；安徽建筑大学引进人才和博士启动基金项目（2023QDZ14） |
| 作者 | 单位 | | 李早 | School of Architecture and Planning,Anhui Jianzhu University,Hefei 230601,China | | 徐梦薇 | School of Architecture and Planning,Anhui Jianzhu University,Hefei 230601,China | | 汪强 | School of Architecture and Planning,Anhui Jianzhu University,Hefei 230601,China | | 蒋文倩 | School of Architecture and Planning,Anhui Jianzhu University,Hefei 230601,China |
|
| 摘要点击次数: 10 |
| 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，将其作为稳定扩散模型的生成基础；随后训练四个低秩适应模型（LoRA），开展传统村落立面生成实验。研究对生成图像开展主客观评价：客观评价采用弗雷歇初始距离（FID）分数，对比生成图像与真实图像，判定生成样本质量；主观评价采用满意度问卷调查法，问卷取得了正向反馈。运用因子分析对问卷数据进行归纳，提取出四类影响因子，并构建多元线性回归方程展开深入分析。结果表明，形态美学因子对整体满意度的影响最大。依托皖南民居立面要素特征，利用模型自动生成建筑立面，可为村落民居立面设计提供一条高效优质的技术路径。与此同时，这套主客观相结合的评价方法，也为解决生成图像的质量判定问题提供了可行思路。 |
| 英文摘要: |
| Under the context of rural revitalization in the new era, the preservation and renewal of traditional dwellings in southern Anhui have become a significant research topic. By analyzing the characteristic elements of Huizhou-style architectural façades as the foundation for generating a stable diffusion model, four low-rank adaptation（LoRA）models were trained to conduct experiments on generating façades for traditional villages. The generated images were evaluated both objectively and subjectively: objective evaluation used Fréchet Inception Distance（FID）scores to compare the quality of generated images with real ones, while subjective evaluation employed a questionnaire-based satisfaction assessment, yielding positive feedback. Factor analysis was applied to summarize questionnaire data, identifying four key influencing factors, which were further analyzed through multiple linear regression. Results indicate that morphological aesthetics exert the greatest influence on overall satisfaction. The model, based on the façade characteristics of southern Anhui, enables automated generation of building façades, offering an efficient and high-quality technical approach for village dwelling design. Meanwhile, the combined use of objective and subjective evaluation methods provides a feasible solution for assessing the quality of generated images. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
