基于稳定扩散模型的中小尺度风景园林空间布局方案生成探索

Exploration of Small and Medium Scale Landscape Architecture Spatial Layout Scheme Generation Based on Stable Diffusion Models

人工智能算法如何有效地提取风景园林设计特征并应用于实际规划设计中是一个值得探讨的问题。已有的研究常集中于利用生成对抗网络构建生成设计技术流程,且在研究过程中也体现出了该算法应用于生成设计的问题,而对于使用稳定扩散模型(Stable Diffusion Models)应用于风景园林方案生成的方法研究则相对较少。鉴于此,文章以特殊标注的平面方案图标签建立用于深度学习的数据集,采用稳定扩散模型算法,通过算法生成风景园林设计方案,并进一步从生成方案对于标注的特征提取能力、方案的合理性等方面评价模型的方案生成结果,探索该训练模型是否有潜力被应用于中小尺度的空间方案设计中。

Artificial intelligence algorithms how to effectively extract landscape garden design features and apply them to the actual planning and design is a problem worth exploring,the existing research often focuses on the use of generative adversarial network to build generative design technology process,and in the research process also reflects the algorithm applied to the generation of design issues,and for the use of Stable Diffusion Models applied to the generation of landscape garden programs is relatively little research.Stable Diffusion Models applied to the generation of landscape architecture programs is relatively little research.In view of this,this paper establishes a dataset for deep learning with specially labeled planar plan labels,adopts the Stable Diffusion Models algorithm,generates landscape garden design schemes through algorithm training,and further evaluates the scheme generation results of the model in terms of the feature extraction ability of the generated scheme for the labeling,the reasonableness of the scheme,etc.,to explore whether the training model has the potential to be applied to the scheme design of small and medium scales of space.We also evaluate the results of the model in terms of its ability to extract labeled features and its reasonableness to explore whether the training model has the potential to be applied to the design of small and medium scale spaces.