ImageSearch

Purpose

Given an input, finds similar image using cosine similarity.

How it was built

The data was turned into a feature embedding vector through a library called img2vec. Then, the vectors were pushed into a vector database (namely pinecone). Then, using pinecone, we query for the most similar match using cosine similarity.

How to use it

• Create a new pinecone index.
• Add images to the imgs/ folder
• Make sure to set the environment variable PINECONE_APIKEY to your pinecone API token.
• Run updateDB.py
• Run main.py. Update the code to search similarities for an image of your choice.