Semantic picture search with MongoDB

Introduction

Should be used together with mdb-search repo, but can be used as a processing backend standalone.

This will download each movie's poster image and generate embeddings with clip-ViT-B-32. Those pictures and their embeddings are stored in MongoDB.

Set-up environment

You need python3 and pip.

python3 --version
python3 -m ensurepip --upgrade
pip3 install -r requirements.txt

You need a MongoDB Atlas cluster. This can be a free cluster, created on cloud.mongodb.com. Ensure database access and network access allow you to make a connection to the database. Note free clusters have a size and performance limitation, feel free to run this on a small paid cluster with lots more data.

You need to set some local environment variables, this can be local .env file

MDB_CONN=<YOUR MongoDB Atlas connection string>
DB="sample_mflix"
COLL="embedded_movies"

Preparing the data

Option 1: Load from a backup

Unrar the dump files from the project directory, run the mongorestore to restore the sample_mflix and embedded_movies collection. You will need the MongoDB command line database tools for this.

mongorestore --uri="mongodb+srv://..."

Option 2: Create the vector search embeddings yourself on any collection

Load the sample dataset from MongoDB Atlas. It's available with the ... in the cluster view. When that's done, run the util.py utility.

python util.py

Attention: It will download pictures for movies and for those pictures create the embeddings using the clip-ViT-B-32 model. This might take a while. Could be optimized with multithreading locally and batch insert_manys. It's normal that you run out of disk space around movie 3000 on the free tier, you can ignore the message.