Housing Passports V2:
Dominica.#

Development Seed has partnered with the World Bank to advance a groundbreaking method for extracting crucial information about building structures using open street-level imagery. Building upon the success of earlier iterations, such as the initiative undertaken from 2019 to 2020 as part of the Global Program for Resilient Housing (GPRH) - detailed in this blog post, particularly notable for its effectiveness in Colombia, where it significantly expedited the identification of vulnerable housing and minimized the need for lengthy field visits and manual inspections. The World Bank is now extending the application of this methodology to the Caribbean region.

The Housing Passports project entails the systematic collection of high-resolution street-level imagery through comprehensive neighborhood surveys. This imagery serves as the foundation for generating annotations based on key characteristics of interest, including building completeness, condition, material, use, and security. Subsequently, machine learning models are trained to automatically identify these attributes. Housing Passports encompasses a robust workflow involving annotation, model training, and processing of machine learning outputs to align them with building footprints, facilitating comprehensive analysis and decision-making.

This website serves as documentation for the project code which lives at developmentseed/housing-passports-v2. Please navigate through the pages on the left as needed. The order follows the sequence of the project workflow:

  1. Dataset creation: see this labeling guide https://devseed.com/housing-passports-labeling/ and the guide for obtaining data from Mapillary and processing it into left/right side images: developmentseed/housing-passports-v2

  2. Building detection model training

  3. Building detection model evaluation

  4. Building classification model training

  5. Building classification model evaluation

  6. Building detection and classification inference

  7. Triangulation (attribution of building footprints)

\(~\)

Built with Jupyter Book 2.0 tool set, as part of the ExecutableBookProject.

ISBN: *(tbd)