Meeting 2 - Notes #7
Description
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