Product Recommendations- Show and Tell

Hi, I always wondered how ecommerce websites built their recommendation engines that recommend similar models to us. So, I took it as a challenge to build myself a small model.
Here’s a small static deployed model for product recommendations, ResNet50 was used and the transfer learning technique is used to train the model.
The deployed model is just a small prototype but I liked how accurate it was! The similarity scores of the products are also listed.

Here is the working prototype on heroku : Product-recommendations


For more details and the model with more than 2000+ images used for training: Click here


Real cool seeing ResNet used in real-time in a Dash app! I’m actually surprised it runs so fast on CPUs :grin:

In terms of modeling, did you train it as a classifier (cross entropy) or using a similarity loss?