Streetscape Analysis with Generative AI (SAGAI)

What is SAGAI?

SAGAI is an open-source, zero-shot workflow designed to score and map street-level urban environments using generative vision-language models. With no need for labeled datasets or deep learning expertise, SAGAI lets users analyze streetscapes through prompts using nothing more than a bounding box. SAGAI (Streetscape Analysis with Generative AI) is also a modular geospatial pipeline powered by the LLaVA model (Large Language and Vision Assistant). It enables prompt-based analysis of Google Street View imagery and is ideal for researchers, planners, and students who need to quickly assess urban scenes and create a map of a given study area.

You can:

• Classify environments (e.g., urban vs. rural)
• Detect features (e.g., storefronts, sidewalks)
• Estimate physical attributes (e.g., sidewalk width)
• Adapt prompts for new tasks

All you need is a location—no pretraining, no fine-tuning, and no annotation required.

How It Works

SAGAI runs in four steps:

1. Street Sampling
Using OpenStreetMap, SAGAI automatically generates sampling points across pedestrian-accessible streets within your selected bounding box.

2. Image Downloading
It fetches Google Street View images at each sampling point, in four directions, using the Google Maps API.

3. Prompt-based Scoring
Images are processed using LLaVA v1.6 (Mistral-7B), which assigns scores or labels based on your input prompt (e.g., “How wide is the sidewalk?”).

4. Aggregation and Mapping
Scores are mapped to streets and points using GeoPandas and visualized as thematic layers.

Each module runs in Google Colab, requiring no installation.

What You Can Use It For

SAGAI is flexible and supports a wide range of geospatial AI applications:

• Urban vs. Rural Scene Detection
• Storefront Presence Estimation
• Sidewalk Width Measurement
• Prompt-based Custom Visual Queries

Why SAGAI?

• 100% open-source and Colab-compatible
• Fully modular and adaptable
• Runs in-browser with no GPU needed
• No pretraining or ML expertise required

Get Started

You can explore the project and run it on your own city via GitHub:

GitHub Repository: github.com/perezjoan/SAGAI
Documentation: Full instructions are provided in each module’s notebook.

Partnership & Support

SAGAI was developed as part of the EMC2 project, co-funded by ANR (France), FFG (Austria), MUR (Italy), Vinnova (Sweden) and the European Commission (Driving Urban Transition Partnership).

Citations & Publications

If you use SAGAI in your work, please cite:

Perez, J. & Fusco, G. (2025). Streetscape Analysis with Generative AI (SAGAI): Vision-Language Assessment and Mapping of Urban Scenes. Geomatica, Volume 77, Issue 2, 100063.
DOI: 10.1016/j.geomat.2025.100063

Assessing urban scenes for the 15-minute city through SAGAI (Streetscape Analysis with Generative AI) – Presentation – 24th European Colloquium on Theorietical and Quantitative Geography