ChromaDB Multimodal Search

This application demonstrates how to use ChromaDB to create a multimodal search system. It allows you to find similar images based on a query image or a text query.

This guide prioritizes using Docker as it's the simplest and most reliable way to run the application, avoiding any local Python environment issues.

Prerequisites

• Docker

Usage (Docker - Recommended)

This workflow uses Docker volumes to store your vector database (my_vectordb) on your host machine. This ensures your data persists even after the container stops.

1. Clone the Repository

git clone git@github.com:trnqeu/chromadb-multimodal.git
cd chromadb-multimodal

2. Build the Docker Image

docker build -t chromadb-multimodal .

3. Prepare Local Directories

You need a local directory to store the persistent database and a directory containing the images you want to upload.

# 1. Create a directory for the persistent database
mkdir ./my_local_vectordb

# 2. Make note of your image directory's absolute path
# For this guide, we'll use /path/to/your/images as a placeholder.
# Example: /home/user/Pictures/my-dataset

4. Run Setup Scripts

Run the setup scripts inside the container, mounting your local directories as volumes.

a. Create the Collection:

This command runs the create_collection.py script, which will create the multimodal_collection inside your ./data folder.

docker run --rm \
  -v ./my_local_vectordb:/app/my_vectordb \
  chromadb-multimodal \
  python create_collection.py

b. Upload Your Images:

This command runs the upload_images.py script. It mounts your local database (so it can add to it) and your local images (so it can read them).

Note: Replace /path/to/your/images with the absolute path to your images.

docker run --rm \
  -v ./my_local_vectordb:/app/my_vectordb \
  -v ./data:/data \
  chromadb-multimodal \
  python upload_images.py /data

5. Run the Application

Finally, run the main Gradio application. It will connect to the persistent database you just populated.

docker run --rm -p 7860:7860 \
  -v ./my_local_vectordb:/app/my_vectordb \
  -v ./data:/data \
  chromadb-multimodal

You can now access the web interface at http://localhost:7860.

---

Managing the Collection

Delete the Collection

To delete all data and start over, run the delete_collection.py script using the same volume mount:

docker run --rm \
  -v ./my_local_vectordb:/app/my_vectordb \
  chromadb-multimodal \
  python delete_collection.py

---

Local Development (Optional)

If you prefer to run the application locally without Docker, you can use pipenv.

Prerequisites (Local)

• Python 3.12+
• pipenv (Recommended installation: pipx install pipenv)

1. Install Dependencies

pipenv install

2. Run the Scripts

a. Create the Collection:

pipenv run python create_collection.py

This will create a my_vectordb directory in your project folder.

b. Upload Images:

pipenv run python upload_images.py /path/to/your/images

c. Run the App:

pipenv run python app.py

This will start the Gradio web interface at http://127.0.0.1:7860.