Semantic Image Search with Redis

This demo show how implement a semantic image search using the entire contents of the National Gallery of Art. It has all the code needed to load the images, create embeddings from them, and store them in Redis as well as to then use these embeddings to search for similar images.

This README instructs you on how to get this demo up and running.

Requirements

You'll need the following bits and bobs installed to get this demo up and running:

• A recent version of Docker—to run Redis 8
• Python 3.12, and not Python 3.13—to run the art-deco-server and, very optionally, the art-deco-loader.
• A recent version of Node.js, I used Node.js v22.15—to run the art-deco-client

Downloading and installing these fine things is left as an exercise for the end user.

Running the demo

You will need to install Redis, start the art-deco-server, start the art-deco-client, and optionally run the art-deco-loader.

Install and run Redis

First things first, we need some Redis. I have conveniently provided a docker-compose.yaml file in the redis folder to do this for you. Note that there is a rather large file named dump.rdb in this folder as well. It contains the results of the art-deco-loader having been run and will be loaded by Redis automatically.

cd redis
docker compose up

Run the server

Next we need the server running. In a new terminal, set up the server's Python environment:

cd art-deco-server
python3.12 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Once it is set up you can run it with:

fastapi dev main.py

You can see if it is running by hitting the root server at http://localhost:8000 with your browser. You should get back:

{
  "status": "OK"
}

Run the client

To run the client, you'll need to set up your Node.js environment. In yet another terminal, run the following:

cd art-deco-client
npm install

Then run the dev target in packages.json:

npm run dev

Run the loader (not recommended)

The loader has already been run for you so you don't really need to do anything here. However, if you want to run it, you'll need to set up your Python environment. So, in a new terminal, run the following:

cd art-deco-loader
python3.12 -m venv venv
pip install -r requirements.txt
source venv/bin/activate

From here you can then run it. It's hard-coded to read the file artwork-data.csv and to call the server on localhost and port 8000. Change it as you need and then run it:

python loader.py

Note that to load the entire file will take some time as it needs to download about 120,000 images and create embeddings for this. It took several hours on my Intel-based MacBoook Pro.

Using the demo

In the left pane, you can upload an image with the "Choose File" button. There are some images to choose from in the images folder or just pick something from your machine. Nothing is stored so you can use whatever you like.

Then, click the "Find Artwork" button and the applicaiton will ask Redis to find the four-closest matches from the 120,000 embeddings stored, download the images from the National Gallery of Art, and present them to you.