Meeting 3 - Notes

[machov]

Group 1 - Signage

• Amber is working to obtain images from particular neighborhoods.
• Varun suggested not to spend too much time automating the tasks.
• Duke used LION shape files to access intersections.
• Duke filtered out the images using PUMA subdirectories.
• Duke and Karan agreed to train an independent model and have specific metrics for each neighborhood.

Group 2 - Taxis

• Varun explained how to import OSC images:
  • Use the openstreetcam.org/ + name, eg.: http://openstreetcam.org/details/1307235/9122/track-info
  • OpenStreetCam cannot provide the images using location as an input, however it is possible to extract images by clicking into the location in the map of the website. See repo: https://github.com/Streets-Data-Collaborative/OpenStreetCam-GeoParsing-Tool.
  • Click the location and reveals all the track images. Each track is a collection of images. Image ID is reveled in the url. Then pull the ID associated with that city.
• Varun requested to be updated about the results of OSC ELT.
• Varun encouraged to include Green taxis and any other important vehicle types.

General

• For given NYC neighborhood, extract all addresses in that neighborhood, and using Google Street View, extract images.
  • For example get all addresses from 1601- 81st St. Brooklyn, NY 11214 and 1605- 81st St Brooklyn, NY 11214.
• Varun encouraged each group to answer the following questions:
  • Group 1: What types of Street signs were observed in Bensonhurst using Google Street View Imagery?
  • Group 2: How many yellow taxi were observed in Bensonhurst using Google Street View Imagery?
• Varun suggested to read the Economist article "Large Economic Gains Can Come From Mundane Improvements In Policy"
• Varun explained that this project intends to be implemented in other regions/cities using Open Street Camera (OSC) open-source images and will provide a street quality "accounting" service.

Other

• Urwa explained Computer Vision concepts and Varun encouraged Urwa to write an Issue about how Computer Vision works.
• Varun commented events about Civic Tech at a Microsoft sponsored tank called Data & Society. They hold weekly lunch talks "Data Bites" in Manhattan.

Next Meeting

• Tentative: next Sunday 13th 2019, 1:00pm.
• We will discuss challenges to answer the above questions and discuss preliminary results.
• Each teammate to test the models with three images.

Attendees:
Varun, Junjie, Duke, Amber, Urwa, Manrique

[vr00n]

Thanks for the detailed scribing @mv1742 ! These are great.

@muaz-urwa I really liked your explanations of Transfer Learning. Please write them down on this repo via an issue and tag as tutorial. Do provide some helpful resources.

Here's something that helped me grok NN's better.
https://experiments.withgoogle.com/what-neural-nets-see