A deep learning hackathon project that uses computer vision to identify and tag products.
PyTorch, Python, Node.js, Express
While interning at Wish, my friend Jeffrey and I competed in an internal Hackathon focused around improving any aspect or process within the company. Inspired by the machine learning projects around us, we decided to create an automatic product tagger to help handle automatically tag the sizable volume of products that Wish handles each day. A neural network was trained using PyTorch such that computer vision could be used to analyze product images that were uploaded to Wish and automatically categorize them into predefined categories. Prior to this project, we both had no experience with machine learning and so this project was extremely fun to work on, especially since our final result was fully functioning.
Highlights from developing this project include:
- Setting up P2 instances on AWS to train our neural network
- Gathering a large enough data set to train our model on for the various predefined categories that it was supposed to be able to identify
- Implementing the trained model prediction service using a Python server and having it communicate with the Node.js server powering our front-end
- Rewriting our demo front-end in 15 minutes because we decided that we were unsatisfied with our existing demo right before we were supposed to present
- Understanding more about machine learning, computer vision, and neural networks