Dataset of Generated Sustainable Streetscape with Generative AI

Sustainable Streetscape with Generative AI This repository accompanies the study "Interpreting ‘sustainable’ streetscapes with generative AI: Context-rich vs. generic prompting".It documents the dataset, experimental design, and prompting strategies used to explore how generative AI interprets and visualizes sustainability in urban streetscapes. Street-View Imagery (SVI) Source: Google Street View (GSV). Selection: 100 unique locations across Jakarta, including Sudirman CBD, Gelora Bung Karno Stadium, and Blok M. Specifications: Images were extracted, ensuring coverage of the streetscape. Resolution: All SVIs stored at 640 × 640 px. To maintain comparability, AI-generated images (originally 1024 × 1024) were downscaled to 640 × 640 px. Prompting Strategy We tested two different prompting strategies to assess how contextual detail affects AI-generated representations of sustainable streetscapes. With Context "Generate a realistic in size of 640x640, high-resolution transformation of the attached street-view image, reimagined as a sustainable streetscape. Preserve the core geometry and perspective of the original view while enhancing it with design interventions aligned with environmental, social, and economic sustainability goals.The image is a street-view image was taken in Jakarta. The street view captures the typical urban character.Enhance the streetscape by widening sidewalks and replacing standard paving with permeable, locally sourced materials that integrate seamlessly with the existing layout. Introduce continuous rows of native trees and landscaped buffers to create shade, reduce urban heat, and improve air quality. Bicycle lanes are added—safely separated from traffic—to support active mobility, while street furniture made from recycled materials provides spaces for resting and social interaction. LED or solar-powered lighting is embedded along pedestrian paths for energy efficiency and safety. Pocket parks or micro public spaces break up the hardscape, encouraging community engagement and biophilic experience. Where appropriate, smart elements like solar-powered benches or environmental sensors may be subtly integrated, contributing to a forward-thinking, inclusive, and ecologically responsive urban street environment." Without Context "Generate a realistic in size of 640x640 transformation of the attached street-view image with the goal of making the streetscape appear more sustainable. Maintain the original structure and urban layout, but creatively reinterpret the scene through a sustainability-focused lens. Emphasize visual realism and plausible urban design improvements.The image is a street-view photograph taken in Jakarta. The street view captures the typical urban character." Purpose This repository is meant to share datasets and methods used for experimenting with generative AI in urban sustainability research.By releasing both the raw street-view baselines and the prompting strategies, we hope to provide a foundation for further exploration of how AI models基于生成式人工智能（Generative AI）的可持续街道景观 本仓库配套论文《基于生成式人工智能解读“可持续”街道景观：富语境提示与通用提示对比》，其中记录了用于探究生成式人工智能如何解读并可视化城市街道景观可持续性的数据集、实验设计与提示策略。 ## 街景影像（Street-View Imagery, SVI） 数据来源：谷歌街景（Google Street View, GSV）。 筛选范围：雅加达市内100处独特点位，涵盖苏迪尔曼中央商务区、格罗拉蓬卡诺体育场以及布洛克M街区。 影像规范：提取影像时确保覆盖完整街道景观。 分辨率：所有街景影像均存储为640 × 640像素。 为保证可比性，将原本分辨率为1024 × 1024像素的AI生成影像统一下采样至640 × 640像素。 ## 提示策略 我们测试了两种不同的提示策略，以评估语境细节对生成式人工智能生成的可持续街道景观表征的影响。 ### 带语境提示组 "生成尺寸为640×640像素的写实风格转换图像，将附随的街景影像重塑为可持续街道景观。在保留原始视角与核心几何结构的基础上，通过符合环境、社会与经济可持续发展目标的设计干预对场景进行优化。本影像拍摄于雅加达，展现了典型的城市街道风貌。具体优化措施包括：拓宽人行道并将标准铺装替换为可渗透、本地采购的材料，确保与现有布局无缝融合；增设连续排列的本土乔木与景观缓冲带，以提供树荫、缓解城市热岛效应并改善空气质量；增设与车流物理隔离的自行车专用道，支持主动式出行；采用回收材料制作街道家具，为休憩与社交互动提供空间；在步行路径沿线嵌入LED或太阳能照明设施，兼顾节能与安全需求；增设口袋公园或微型公共空间以打破硬质铺装界面，促进社区参与并营造亲自然体验；可视场景中可适度集成太阳能长椅、环境传感器等智能设施，打造兼具前瞻性、包容性与生态适应性的城市街道环境。" ### 无语境提示组 "生成尺寸为640×640像素的写实风格转换图像，基于附随的街景影像，将街道景观优化为更具可持续性的形态。保留原始场景结构与城市布局，以可持续发展视角对场景进行创新性重构，注重视觉写实性与合理的城市设计优化。本影像拍摄于雅加达，展现了典型的城市街道风貌。" ## 研究目的 本仓库旨在分享用于城市可持续性研究中生成式人工智能实验的数据集与方法。通过公开原始街景基准影像与提示策略，我们希望为进一步探究人工智能模型如何