P&G Click Predict Innovate

Visibility norms of products in retail stores dictate how and where they should be positioned in the store, to achieve optimum visibility and increased sales. Today the process for validating visibility norms in Traditional Trade is based on audit activities conducted in a subset of stores via external agencies leading to lack of timely actionable insights. Manual audit within a store via a Sales Rep is also a biased view due to same individual doing audit and taking action. Using object detection, create a system to track and validate visibility norms in all P&G Stores in a fast, cost-effective, unbiased, and automated way.

You are expected to create deep learning models that detect multiple products from shelf images and classify them into different categories as given in the dataset. Detected products should be categorized into 4 categories, mentioned in the dataset. In the dataset, all images are divided in 4 folders, where folder name is the category of all products inside that folder. Bounding box coordinates for each product are also provided, that can be used for detecting product from shelf images. You also need to calculate the quantity of that particular product or product category in the image or shop.

Also make sure that you develop your solution using deep learning. Submissions implementing OpenCV’s built-in object detection will not be considered.

The data set for Object Detection can be accessed here. We have added to the data set given last week. Another set of annotated images can be found here. Please download both.

Sales Representatives of P&G go from store to store, asking retailers to stock up on P&G products basis availability of different categories of products. Average call time of a Sales Rep within a store is 20 mins. Today a majority of the Sales Reps’ time is spent in asking retailers about availability of products and reviewing what to sell, as opposed to executing the sale . Any intelligent recommendations he does get on his Selling Device are based on historical sales data. There is no consideration for product availability within the store at that time, which wastes a lot of time.

You are expected to create a product recommendation system using machine learning and the given dataset. You are free to use both content based filtering or collaborative filtering. Given dataset contains a CSV file with user_id, product_id and rating of product. Based on that, you have to recommend products to users. You are also free to use any other publicly available datasets, if you want.

If your model is selected, you will later get to work with P&G’s dataset as part of the project and build a recommendation system relevant to P&G’s products.

Please click here to access the example data set. Column 1 is the User ID, Column 3 is Rating, and Column 4 is Product ID. Column 2 is irrelevant.

Create any image recognition based AI system that you feel will benefit P&G's retail strategy in the emerging markets.