StreetLens: Enabling Human-Centered AI Agents for Neighborhood Assessment from Street View Imagery
This is the official repository of StreetLens.
StreetLens is a human-centered, researcher-configurable workflow for scalable neighborhood environmental assessments. It enables VLMs to mimic trained human coders by:
- Domain-informed analysis: Focuses on questions derived from established protocols.
- Scalable image retrieval and annotation: Automatically retrieves relevant street view imagery (SVI) and generates semantic annotations, ranging from objective features (e.g., number of cars) to subjective perceptions (e.g., sense of disorder).
- Flexible integration of expertise and data: Allows researchers to define the VLM’s role through domain-informed prompting and incorporate prior survey data for robust assessments across diverse contexts.
- Demo: https://knowledge-computing.github.io/projects/streetlens
- ACM SIGSPATIAL GeoHCC'25 Workshop Paper: https://dl.acm.org/doi/10.1145/3764917.3771334
- arXiv Paper: https://arxiv.org/abs/2506.14670
Below is an overview of the StreetLens workflow along with input examples from a case study:
Figure: Input examples from a case study and system architecture of StreetLens showing the flow of VLM-based neighborhood assessment.
We provide two Google Colab notebooks that can be run with a free GPU quota:
1_data_exploration.ipynb– Explore the input data2_assess_neighborhood_environment.ipynb– Run neighborhood environment assessment
For more details on the methodology, see the paper: StreetLens: Enabling Human-Centered AI Agents for Neighborhood Assessment from Street View Imagery
If you find this work useful, please cite it using the following BibTeX entry:
@inproceedings{10.1145/3764917.3771334,
author = {Kim, Jina and Jang, Leeje and Chiang, Yao-Yi and Wang, Guanyu and Pasco, Michelle C.},
title = {StreetLens: Enabling Human-Centered AI Agents for Neighborhood Assessment from Street View Imagery},
year = {2025},
isbn = {9798400721809},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3764917.3771334},
doi = {10.1145/3764917.3771334},
booktitle = {Proceedings of the 1st ACM SIGSPATIAL International Workshop on Human-Centered Geospatial Computing},
pages = {15–19},
numpages = {5},
keywords = {automatic workflow, neighborhood environment assessment, vision-language model, prompt engineering, in-context learning},
location = {Minneapolis, MN, USA},
series = {GeoHCC '25}
}