The open-source tools, datasets and platforms that OSCUR develops at New York University, the University of Illinois Chicago and the University of Washington, with their partners. They are grouped by the stage of the urban data lifecycle they support; each lists its code, data and papers.
Finding urban datasets, describing and profiling them, and combining them into analysis-ready data.
NYU
Dataset search engine for data-driven and data-relationship queries
Indexes over 20,000 datasets, including the open-data portals of several US cities. A next generation, built to run as a hosted service over OSCUR’s urban collections, is in development.
NYU
Accurate, readable descriptions of datasets, generated with LLMs
Combines data profiling with large language models, so that datasets with terse or missing metadata become findable by people and by AI agents. Its descriptions improved retrieval by up to 30%.
NYU
Semantic profiling of datasets with LLMs, extending datamart-profiler
Improves the recognition of spatial and temporal attributes, the profiles that dataset search and integration build on.
NYU
LLM-assisted discovery and semantics-aware integration of spatial data
Turns spatial data wrangling into a transparent, node-based workflow driven by context-aware AI agents, with a provenance graph that records how each transformation was derived.
NYU
Spatial enrichment of urban layers with Scikit-Learn-style pipelines
Links urban datasets to street networks, sidewalks, city features and administrative boundaries in reproducible, shareable pipelines.
NYU
Curated, profiled urban datasets for New York City
Sources across infrastructure, traveler behavior and safety, and urban context, documented and profiled for search, and published as analysis-ready datasets on Hugging Face.
Building analyses as dataflows that can be shared, reproduced and traced back to their steps.
UIC
Dataflow framework for collaborative urban visual analytics
Organized around extensible data, node and agent catalogs, and used in UIC courses and with Chicago Public Schools.
UIC
Dataflows for human-AI alignment in urban visual analytics
Lets analysts state their intent for a whole dataflow, a node or a parameter, with explanations and the provenance of their interactions.
NYU
VisTrails workflows and provenance, reimplemented in Julia
Action-based provenance with branching and merging, workflow replay and diff, notebook-defined workflows and git-native versioning, and it opens existing VisTrails files.
In development
Grammars, toolkits and systems for exploring urban data on maps, in 3D and along streets.
NYU · UIC · UFF
Serverless toolkit for prototyping urban visual analytics in the browser
Spatial data management, GPU computation, 3D maps and charts in one lightweight library, with abstractions designed to work well with AI-assisted coding.
UIC
Grammar-based framework for urban visual analytics
A declarative grammar that integrates physical and thematic urban layers in 2D and 3D maps.
UIC
Declarative grammar for street-overlaid visualization of multivariate data
UIC
Knowledge base of the building blocks of visual analytics systems
A structured, machine-readable blueprint of how systems compose data processing, analysis, visualization and interaction, extracted from papers with LLMs.
UIC
Neural fields for view computation and data exploration in 3D cities
Encodes a 3D urban environment as an implicit field that answers direct and inverse queries about visibility, sunlight and the impact of new buildings.
UIC · UC Berkeley
Interactive topology repair of automatically extracted sidewalk networks
Detects disconnected components, dangling edges, missed connections and malformed intersections in networks from pipelines such as Tile2Net, and supports their repair with real-time quality feedback.
UIC
Scenario-oriented urban toolkit for decision support
NYU · FGV
Event and scene retrieval in long urban intersection videos
Combines retrieval-augmented generation, taxonomy-aware entity extraction and video grounding to make hours of street footage searchable in natural language.
Computer vision and AI pipelines that turn street-level and aerial imagery into urban data.
UW
Curb ramp detection bootstrapped from open government data
Turns municipal curb ramp inventories into labels on more than 210,000 Street View panoramas and trains a detector on them. It runs in Project Sidewalk, labeling curb ramps for people to validate.
UW
Detecting disability parking and its quality from aerial imagery
UW
Measuring socioeconomic bias in Street View coverage
A reproducible pipeline and dashboard that relate the age of Street View imagery to census indicators, so that analyses can account for coverage bias.
UW
Personalized, context-sensitive bike routing from open data and VLMs
UW
Detecting and exploring urban art at city scale with VLMs
A five-stage VLM pipeline over Street View panoramas, and a screen-reader-accessible tool for finding, filtering and touring public art.
Code and benchmark to come
UW · U Tokyo
Predicting navigation landmarks with VLMs, shown in mixed reality
Comes with an open dataset of more than 3,000 360° street-view images with gaze and landmark annotations.