E-Commerce Object Recognizer Project

Date:

A web service which helps people find the name of the products by just taking a photo with their cell phone.

The tools used:

  • YOLO - Real Time Object Detection
  • TensorFlow For Poets
  • Flask and Python Libs
  • Web Driver

Methodology:

  • The object with highest accuracy is detected, extracted and cropped from the image.
  • This cropped image is now given as an input to the classifier that we trained.
  • The classifier then predicts the brand of the image.
  • We used a pre-trained model from image-net and we retrained the classification layer of our model to predict and classify a few brands.
  • We used the Python Flask framework as our web server and HTML for the web app.

Future work:

  • Improve training using a bigger dataset and.
  • Try real time object detection and predictiction through camera.
  • Accurate predictions would further enable automated web crawling and integration of price comparison systems.