Home

Welcome to the rAPIdtools wiki. rAPIdtools is a Python toolkit from the NHERI RAPID Facility for running large-scale AI inference and localization on post-disaster geospatial data: it connects orthomosaics, footprints and street-level imagery to vision-language models and returns GIS-ready results.

What it does

• Ingests geospatial data at scale. PhysicalAssetCollection loads GeoJSON and ESRI shapefiles, filters by attribute or region, and exports back to GeoJSON, shapefile or a pandas DataFrame.
• Pulls imagery for every asset. Crops from local orthomosaics, keyless Bing and Google satellite tiles, Google Street View panoramas and Mapillary street-level images, each attached to the asset it depicts.
• Runs vision-language models locally or through APIs. One AssetAnalyzer works with local Hugging Face checkpoints (Gemma, Llama, Qwen, Muse Glimmer and others) and hosted providers (OpenAI, Gemini, Claude, Muse Spark, Qwen), with batching, rate-limit handling and cancellation built in.
• Finds and regularizes features. SAM 3 detection over orthomosaics, building and road regularizers that turn pixel masks into clean geometries, and street-level object discovery from Mapillary detections with camera-geometry localization.
• Ships a browser app. rapidtools-gui runs the whole workflow without code and can be shared with a team on one GPU machine.

Where to start

If you want to...	Read
Install the package and set up credentials	Installation
Run your first pipeline in a few lines	Quickstart
Understand assets, images, pipelines and models	Core Concepts
Pick a model and know what it needs	Supported Models
Work from orthomosaics or satellite tiles	Aerial Imagery Workflows
Work from Street View or Mapillary imagery	Street-Level Imagery Workflows
Tune prompts, generation settings and rate limits	Vision-Language Models
Save, reload and exchange results	Data Formats and Export
Use or share the browser app	Graphical Interface, Deployment
Fix an error	Troubleshooting and FAQ
Change the code	Contributing and Development

Worked, runnable scripts are indexed on the Examples page.

How this wiki relates to the documentation site

The documentation site is generated from the source tree and is the reference for every class and function, with one page per object under its API reference. This wiki is the narrative companion: how the pieces fit together, which options matter in practice, and what to do when something goes wrong. When the two disagree, the documentation site built from main is authoritative; please open an issue so the wiki can be corrected.

Project links

• Source and issues: github.com/RAPID-Facility/rAPIdtools
• Feedback and feature requests: request a feature, report a bug or send feedback / ask a question; the documentation's feedback page explains what happens next; without a GitHub account, email uwrapid@uwrapid.org
• Releases on PyPI: pypi.org/project/rapidtools
• Changelog: CHANGELOG.md
• How to cite: CITATION.cff
• License: BSD-3-Clause