This capstone project was executed for the 'Python for AI & Data Science' training program.
As the capstone project for Batch-1, an individual was assigned a product category. I was assigned ‘Coca Cola’. The scope included the following:
- Capture 50-100 images of the assigned product placed on supermarket shelves.
- Annotate the images and split into 3 datasets: train, validate & test sets.
- Use Google Collab to train Detectron 2 model.
- Predict on images & extract image parameters such as digital coordinatres, logical coordinates, image label & confidence level.
The main implementation of the project is to recognize the products on supermarket shelves. Various cameras can be placed that give real-time imagery to the model. The model can therefore, be able to count the objects placed on a particular shelf. This application will be very useful in:
- Real-time Inventory Assessment Maintaining the level of inventory at big supermarkets is indeed a challenging problem. By using AI, a real-time estimate of the products available on shelves can be made. This will help optimize the supply chain of the super-market and save costs.
- Theft-detection Although, CCTV cameras are available in supermarkets, supermarket shrink (loss of product inventory) & shop-lifting is still prevalent. A spring 2013 study by the University of Florida showed a 2.5 percent supermarket shrinkage rate, which was more than double than the 1.1 percent rate for all retailers. Shoplifting accounted for 36 percent of all theft-related shrinkage.
For access to the dataset please contact [email protected]