ARGO internship project to use http://customvision.ai to detect features from US Street imager
Small improvements in "Maintenance, management of state assets and public-sector accounting", according to a recent Economist op-ed could lead to large economic gains for cities, states, and entire nations.
We believe that local and state governments should be capable of auditing the urban environment using common-sense digital tools and leveraging advances in machine learning, responsibly towards the next-generation of Maintenance, management of state assets and public-sector accounting.
Using street view imagery, develop a computer vision model that is trained on three specific elements of urban infrastructure, namely:
- Detection of US street signage.
- Road width estimates.
- Detection and counting of Yellow Taxis on NYC streets.
Use http://customvision.ai - an easy-to-use computer vision toolkit developed by Microsoft.
- Read the SQUID Story
- Learn how to use http://customvision.ai to develop quick computer vision applications.
- Learn how to collect street view imagery at a citywide scale using Google Street View API and Open Street Cam - OSC and our OSC-ETL tools.
- Develop and Document simple data specifications for each of the detections.
- Isolate your imagery for model training.
- Upload training imagery to customvision interface.
- Label what you intend to detections from the uploaded imagery.
- Test the accuracy of your trained model.
- Iterate towards better precision and recall.
- Parse results from customvision output into a more readable format.
- Deploy your trained model to "audit" a specific geography. (For consistency, we will choose this area for you).
- Prepare blog to document your journey. What worked, What did not. Ideas for future improvement / applications etc.
ARGO project supervisor: Varun Adibhatla
CUSP Students
- Project scribe: CUSP student who will take notes during check-in and manage the github repo (code,readme, issues)
- Project logistics: CUSP student who ensures meetings are scheduled and handles conference/video calls.