Urban Environment Analysis with Generative AI

The field of urban environment analysis is moving towards leveraging generative AI and street view imagery to estimate urban forest health, assess streetscapes, and create geospatial embeddings. This shift enables scalable and effective management of urban environments, overcoming traditional limitations of instrumented inspection techniques and dedicated deployments. Noteworthy papers include:

  • Streetscape Analysis with Generative AI (SAGAI) which introduces a modular workflow for scoring street-level urban scenes using vision-language models.
  • S2Vec, a novel self-supervised framework for learning geospatial embeddings, yields task-agnostic embeddings that capture local feature characteristics and broader spatial relationships.
  • Unsupervised Urban Land Use Mapping with Street View Contrastive Clustering and a Geographical Prior, which enhances clustering performance with a built-in geographical prior, offering a flexible solution to land use mapping.

Sources

Using street view imagery and deep generative modeling for estimating the health of urban forests

Streetscape Analysis with Generative AI (SAGAI): Vision-Language Assessment and Mapping of Urban Scenes

S2Vec: Self-Supervised Geospatial Embeddings

Unsupervised Urban Land Use Mapping with Street View Contrastive Clustering and a Geographical Prior

Built with on top of