Meeting 2 - Notes

[machov]

Meeting 2

Group 1 - Street Signage

• Group 1 created a customvision model with more than ten categories of signage.
• Karan reported high precision, high accuracy, but low recall for most of the categories.
• Information and bicycle signs had low recall.
• Varun suggested to extract images from street intersection only and use LION files to scrape images from nodes.
• Varun encouraged Karan to write these findings up as an issue or final blog.
• Varun suggested to look at each neighborhood to count. Pick a neighborhood like Clinton or Brooklyn.
• Duke also suggested to test how the model performance changes with image parameters like resolution.

Group 2 - Yellow Taxis

• Group 2 created a customvision model using only one category.
• Urwa uploaded more 9000 images and Junjie ran the customvision model.
• Manrique suggested to also categorize Green taxis.
• Varun suggested to look at Manhattan only to catch more taxis.
• Varun suggested both groups to use OpenStreet Camera images as they are a flat image with high resolution.

Next Meeting

Next meeting to be held next week.
Tentative: Friday January 4th 2019, 1:00pm EST (New York time)

Meeting Attendees:
Varun Adibhatla, Yushi Chen, Junjie Cai, Karan Saini, Sung Hoon Yang, Qiuyao Liu, Manrique Vargas
Could not attend: Urwa Muaz