Evaluation of Synthetic Images Generated for Traditional Residential Architecture Using Few-Shot Approach
Ayşe Berika Varol Malkoçoğlu, Didem BARAN ERGÜL
Samsun University Istanbul Technical University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
In recent years, artificial intelligence-based Text-to-Image (T2I) models have demonstrated a strong capability to generate highly realistic outputs in architectural visualization. However, the extent to which these images accurately and consistently reflect site-specific spatial, cultural, and structural characteristics-particularly within rural and vernacular architectural contexts-has not been sufficiently examined in the existing literature. This study investigates the representational capacity of T2I models with respect to domainspecific architectural attributes by generating synthetic images depicting residential architectures characteristic of distinct regions of Anatolia using the GPT Image$\mathbf{1. 5}$model. The correspondence between the generated images and their intended architectural classes was quantitatively evaluated through a CLIP model enhanced with a few-shot learning approach. In addition to this model-based assessment, a qualitative evaluation was conducted using architecturally grounded criteria, including facade articulation, mass composition, parcel integration, semi-open space configuration, and contextual coherence. This dual-phase evaluation framework enables both objective computational assessment and expert-based scrutiny of architectural content accuracy. The findings indicate that while the GPT Image 1.5 model is largely successful in capturing region-specific architectural characteristics, it still exhibits limitations in accurately representing certain morphology-dependent and contextsensitive components.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程Architecture and Computational Design
Generative Adversarial Networks and Image Synthesis · Urban Design and Spatial Analysis