Urban-WORM: A Multimodal Model-Driven Intelligent Annotation Tool for Crowdsourced Geospatial Data

Urban-WORM is an open-source multimodal inference workflow framework that focuses on generating rich and interpretable automatic annotations for geotagged crowdsourced image data, suitable for urban research, geographic information systems (GIS), and spatial data analysis scenarios. It aims to address the high cost and scalability challenges of traditional manual annotation, providing a user-friendly interface that allows users to build image understanding pipelines without deep knowledge of model details.

With the popularity of smartphones and social media, a large amount of geotagged image data is actively uploaded by users to various platforms. These crowdsourced data contain rich urban spatial information, but extracting valuable insights from them has always been a challenge. Traditional methods rely on manual annotation, which is not only costly but also difficult to scale. Urban-WORM (Workflow Of Reproducible Multimodal Inference) emerged as a solution, providing a user-friendly high-level interface specifically designed to use multimodal large language models to generate rich and meaningful descriptive annotations for geotagged crowdsourced data.

Urban-WORM's design philosophy is "reproducible multimodal inference". It encapsulates complex model invocation processes into a concise workflow interface. Users can quickly build image understanding pipelines without deep knowledge of underlying model details. The tool supports multiple mainstream multimodal models, can process both image content and geospatial metadata simultaneously, and generates structured annotations covering dimensions such as scene description, object recognition, and spatial relationships. This design is particularly suitable for urban researchers, geographic information system (GIS) analysts, and spatial data scientists.

In practical applications, Urban-WORM can serve multiple fields:

• Urban Perception Research: Analyze street view images uploaded by citizens to understand the perceived quality of urban spaces
• Environmental Change Monitoring: Compare geotagged images from different time points to track urban landscape evolution
• Disaster Response Assessment: Quickly process post-disaster crowdsourced images to assist emergency response decision-making
• Cultural Heritage Documentation: Automatically generate detailed descriptive archives for historical buildings and landmark images

A key feature of Urban-WORM is its emphasis on "reproducibility". Each inference process is fully recorded, including the model version used, prompt configuration, and output results, ensuring that research results can be independently verified and reproduced. Additionally, the tool adopts a modular design that supports flexible expansion of new multimodal model backends. Both open-source local models and commercial API services can be integrated into the workflow through a unified interface.

As an open-source project, Urban-WORM is hosted on GitHub and uses a permissive license to encourage community contributions. Project maintainers actively respond to issues and pull requests, forming an active user community. This open collaboration model ensures that the tool can continue to iterate and adapt to the rapid development of multimodal technology.

With the continuous improvement of multimodal large language model capabilities, the potential of tools like Urban-WORM will be further unleashed. Future versions may integrate more advanced visual understanding capabilities, support video sequence analysis, and even combine satellite imagery for larger-scale spatial analysis. For researchers engaged in urban computing and spatial data science, this is an open-source tool worth paying attention to.