Skip to content

Vector database (Weaviate) to build an image search engine powered by a deep neural network

License

Notifications You must be signed in to change notification settings

weaviate-git-bot/Weaviate-Search-Engine

 
 

Repository files navigation

Streamlit logo
Weaviate logo



Weaviate Image Search Engine

Vector database (Weaviate) to build an image search engine powered by a deep neural network. It uses Weaviate optimizer for images with ResNet50 (PyTorch) as vectorizer and retriever. Currently only PyTorch ResNet50 supports CUDA-GPU but in a single-threaded manner. You can also use the keras-based ResNet50 CPU with multi-threaded inference.

Steps to run it

  1. Install Docker (skip this if already installed)
curl -fsSL https://get.docker.com -o get-docker.sh && sh get-docker.sh
  1. Spin up docker compose file:
docker compose up -d
  1. Activate the environment:
pip3 install pipenv && pipenv install && pipenv shell
  1. Launch start.sh script to create Weaviate schema, convert images from img folder to base64 and upload them in Weaviate:
./start.sh
  1. Navigate to URL that streamlit outputs and upload an image. The program will query Weaviate for a vectorized search of similar images. If you are running this in local environment then just navigate at:
http://127.0.0.1:8501/
  1. (Optional) CORS and XsrfProtection Streamlit. Run this if you have Socket Errors while connecting to streamlit app URL:
streamlit run streamlit_app.py --server.enableCORS false --server.enableXsrfProtection false

Images

Feel free to insert your images in img folder and see if the database suggests you similar images.

Weaviate Schema

This is the schema implemented in Weaviate. You can modify the class name, change the vectorizer or insert more properties.

schemaConfig = {
    'class': 'MyImages', # class name for schema config in Weaviate (change it with a custom name for your images)
    'vectorizer': 'img2vec-neural',
    'vectorIndexType': 'hnsw',
    'moduleConfig': {
        'img2vec-neural': {
            'imageFields': [
                'image'
            ]
        }
    },
    'properties': [
        {
            'name': 'image',
            'dataType': ['blob']
        },
        {
            'name': 'text',
            'dataType': ['string']
        }
    ]
}

About

Vector database (Weaviate) to build an image search engine powered by a deep neural network

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 97.0%
  • Shell 3.0%