Finding visual content with semantic meaning is a very important task. It has applications in the field of e-commerce(using camera to shop for a similar furniture), recommendation systems (like PinSage @ Pinterest or discover weekly @ Spotify). In this notebook, I use Annoy to build a rudimentary Reverse Image search system using deep learning. In this system, a sample image formulates a image query. You can find some of the results below and a detailed explanation in the accompanying blog.
Each model is in a column of its own. Query image is the first row the subsequent rows are nearest neighbors for each model
Query image is the first row the subsequent rows are nearest neighbors based on ensemble of all models
