The problem Intelligent Counter Share Analyzer solves

Modern retail chains(DMart, Big Bazaar, etc.) store a variety of
products from different brands. Each brand competes to get better visibility in
the store so a consumer can buy them. The stores charge a premium to the
brands to place their products on the shelf at eye level, or around the corner of the
aisle, or on the billing counter. They also charge a premium to give the majority space
to a particular brand by suppressing the other brand’s visibility. It becomes
extremely difficult for the brands to validate if the stores are complying with the
contract terms by giving actual visibility.
This project is an intelligent counter share analyzer that can analyze video,
images and derive the % of brand presence on a particular section

Challenges we ran into

First, we tried to build an object detection model manually but the accuracy was low so we used transfer learning using ImageAI.
The ImageAI required the dataset annotations to be in pascal VOC format which our data was not in. Hence we had to write a function to convert our annotations CSV in pascal VOC format.
Then faced issues with CUDA because the dataset and the pre-trained model required a lot of RAM, so we decided to use google colab's hosted VM to run the model.