AmazonOracle

Introduction

Given an image of an item, Amazon Oracle returns the predicted category and price of that item, and visually similar items currently being sold on Amazon.

Practical Uses

Find Predicted Price & Category

This app is useful for merchants or consumers who want to check the price of a product.

Find Similar Items

If you are looking at an item, you can use this app to find similar items that leads you to the product page instantly.

Implementation

Data Collection

Data (such as image, price, title, ASIN) was collected from the Amazon.ca marketplace using the Amazon Product Advertising API. (Disclaimer: this was done several days prior to the start of the hackathon). Over 100,000 images were gathered.

Model Building

Using Keras with Tensorflow, we fine-tuned Google's InceptionV3 convolutional neural network, to produce two outputs: a continuous price output, and a discrete category output. A fingerprint is also generated for each image.

Back End

The back-end is built with Django on top of an AWS EC2 instance. This instance does the predictions given a new image.

React-Native

The front-end of the app is built with React-Native. User will upload a photo to our server endpoint through the app with a POST request, then predicted information will be returned for display in the app.

Examples

Home > Predicted Info > Similar Items