Content and Text Based Image Retrieval Search Engine

Context | Getting started | Pipeline | Deployment | Contribution

Context

An image search engine that covers over 2 million annotated images from Open Images Dataset. This search engine enables its users to search using text, images or a combination of both.

Getting started

Prerequisites

Your machine should have installed:

• Docker
• ElasticSearch 7.14.1
• Elastiknn plugin 7.14.1.0
  elasticsearch-plugin install https://github.com/alexklibisz/elastiknn/releases/tag/7.14.1.0

Installation

In order to reproduce the project you have to:

1. Clone the repository.

2. Move to the project directory.

3. Get the data index. This can be done in two methods:

   1. Create a new index
      1. Execute download_data.sh to get the data.
      2. Follow the indexing steps in indexing.ipynb.
   2. Restore our index.
      1. Download the index. https://drive.google.com/file/d/1HVr_i4XYx5uQ67B94pxX83g7ZD1kxOcf/view?usp=sharing.
      2. Go to ElasticSearch/config directory.
      3. open elasticsearch.yml and add the following line:
         path.repo:[path_to_index/elasticsearch-backup]
      4. Reload indexing using PUT request to
         http://localhost:9200/_snapshot/open-images and the following header
         { "type": "fs", "settings":{ "location": "path_to_index/elasticsearch-backup", "compress": true } }
      5. Restore the index using POST request to
         http://localhost:9200/_snapshot/open-images/snapshot_02_12_2021/_restore

4. Run docker-compose up --build Once these steps are done, the web application should start in the browser.

Pipeline

This project is brought down into 4 steps:

Data processing

Data indexing

Data indexing is performed using ElasticSearch python client. This operation is split into two tasks:

1. Indexing text data.
2. Indexing image feature vectors using Elastiknn plugin which enables storing numerical vectors and provides the Approximate Nearest Neighbours queries in ES.

Application development

The web application is developed using:

• ElasticSearch python client: to query elasticsearch index.
• FastAPI : to develop an API to enable querying from the web application.
• Streamlit : to design a frontend for the application where users perform text and image search.

Deployment

Both of the API and frontend of the web application are containerized using Docker and docker-compose.

More about this process is available in this detailed report.

Contribution