Visual and Semantic Similarity with Redis

This demo goes along with the Annoucement of a New Redis Vector Similarity Search

You will experiment with two key applications of Vector Similarity Search application using a realistic dataset:

Semantic Search: Given a sentence check products with semantically similar text in the product keywords
Visual Search: Given a query image, find the Top K most "visually" similar in the catalogue

About the Amazon Product dataset

The CSV product data used in this demo was derived from the "Amazon Berkeley Objects Dataset"

Each row in the CSV file correspond to a product in the original dataset.

Before you start

Install Git LFS
- Make sure you initialize LFS by runing
```
git lfs install
```
Docker with at least 8 GB RAM (for Redis Stack)
Anaconda (for Jupyter notebooks)

Clone the Repo

git clone https://github.com/loriotpiroloriol/vecsim-demo.git

Fire Up Redis Stack

docker run -d --rm --name redis -p 6379:6379 redis/redis-stack-server:latest

Launch the Jupyter Notebooks

jupyter notebook

All notebooks have the option to user either Hash or JSON. Follow either path by running the cells for one option and skipping those for the other.

Step 1: Semantic Similarity - Part I

Open this notebook [http://127.0.0.1:8888/notebooks/SemanticSearch1k.ipynb

Run All Cells and check the outputs

You will generate embeddings for 1,000 products and perform semantic similarity using two indexing methods(HNSW and brute-force)

Step 2: Semantic Similarity - Part II

Open this notebook http://127.0.0.1:8888/notebooks/SemanticSearch100k.ipynb

Run All Cells and check the outputs

You will load ~100k previously-generated embeddings for the first 100,000 products in the dataset. You'll perform semantic similarity on a larger dataset

Step 3: Visual Similarity - Part I

Open this notebook http://127.0.0.1:8888/notebooks/VisualSearch1k.ipynb

NOTE: When you run this cell:

!wget -c https://amazon-berkeley-objects.s3.amazonaws.com/archives/abo-images-small.tar
!tar -xf  abo-images-small.tar -C ./data/

it will take 5-10 minutes (depending on your network). The Jupyter notebook downloads a 3.25GB tar file with product images from the "Amazon Berkeley Objects Dataset". You only need to run this once for both visual similarity notebooks.

Run All Cells and check the outputs

You will generate embeddings for 1,000 product images and perform visual similarity using two indexing methods

Step 4: Visual Similarity - Part II

Open this notebook http://127.0.0.1:8888/notebooks/VisualSearch100k.ipynb

You'll perform visual similarity on a larger dataset using two indexing methods (HNSW and brute-force)

About the Amazon Product data

The dataset used in this demo was derived from the "Amazon Berkeley Objects Dataset"

In particular, each long text field in the product_data.csv was extracted from the original JSON encoded object representing each product.

Thanks to Amazon.com for sharing the original dataset. This includes all product data, images and 3D models under the Creative Commons Attribution-NonCommercial 4.0 International Public License (CC BY-NC 4.0)

Credit to the creators of the dataset: Matthieu Guillaumin Amazon.com Thomas Dideriksen Amazon.com Kenan Deng Amazon.com Himanshu Arora Amazon.com Arnab Dhua Amazon.com Xi (Brian) Zhang Amazon.com Tomas Yago-Vicente Amazon.com Jasmine Collins UC Berkeley Shubham Goel UC Berkeley Jitendra Malik UC Berkeley