aashishyadavally / rubrix

Combined semantic similarity search based visual search engine.

AI Powered Visual Search Engine

The main motivation behind building rubrix was to have a visual search engine completely powered by Artificial Intelligence, tying concepts within the fields of Natural Language Processing and Computer Vision, something we like to call "combined similarity search". Currently rubrix has two main functionalities:

• take in a user input describing an image and retrieve five images that fit that description (image search)
• take in a user uploaded image and retrieve five similar images (reverse-image search)

Please click here to know more details about the architecture and how rubrix works!

rubrix in Action

You can check out some of the images retrieved by rubrix for sample queries here.

Getting Started

This section describes the preqrequisites, and contains instructions, to get the project up and running.

Setup

1. Project Environment

Currently, rubrix works flawlessly on Linux, and can be set up easily with all the prerequisite packages by following these instructions:

1. Download appropriate version of conda for your machine.

2. Install it by running the conda_install.sh file, with the command:

   $ bash conda_install.sh

3. Add conda to bash profile:

   $ source ~/.bashrc

4. Navigate to rubrix/ (top-level directory) and create a conda virtual environment with the included environment.yml file using the following command:

   $ conda env create -f environment.yml

5. Activate the virtual environment with the following command:

   $ conda activate rubrix

6. To install the package with setuptools extras, use the following command in rubrix/ (top-level directory) containing the setup.py file:

   $ pip install .

2A. Data Assets - Setup from Scratch

Once the prerequisites have been installed, follow these instructions to build the project:

1. Navigate to rubrix/index directory.

2. Run the bash script setup.sh with the following command:

   $ bash setup.sh

What does this do?

1. Downloads flickr8k image/captions dataset.
2. Builds and sets up darknet/ within rubrix/index to enable object detection with YOLOv4.
3. Creates assets/index.json file, which essentially is an inverse-image index mapping all the objects YOLOv4 was trained on, to the images containing them.
4. Creates assets/imageEmbeddingLocations.json file, which essentially maps all the images in the database to the sentence embedding vectors generated for each of the captions in the database.
5. Generates feature vectors describing all the images in the database and save it to assets/descriptors directory.

NOTE: The above script can take between 1.5 - 2 hours to complete execution.

2B. Data Assets - Quick Setup

1. Download data assets from this link.
2. Unzip and save the contents in rubrix/assets.
3. All is left is to change the paths in rubrix/assets/index.json and rubrix/assets/imageEmbeddingLocations.json relative to the local machine. This can be done as follows:
   • Ensure corresponding virtual environment is active, or activate with the following command:

     $ conda activate rubrix

   • Launch Python Interpretor in the terminal and run the following code snippet:

     >>> from rubrix.utils import fix_paths_in_index
     >>> path_to_index = <absolute/path/to/rubrix/assets/index.json>
     >>> path_to_emb = <absolute/path/to/rubrix/assets/imageEmbeddingLocations.json>
     >>> fix_paths_in_index(path_to_index, path_to_emb)

4. Navigate to rubrix/rubrix/index directory and run the following bash script:

   $ bash quick_setup.sh

Usage

1. As a package

With the completion of these steps, you should be able to use rubrix.

• For image search, execute the rubrix/query/query_by_text method.
• For reverse image search, execute the rubrix/query/query_by_image_objects method.

You can also follow a working example for this here.

2. Local Web Application

An alternative is to use rubrix as an application on web browser.

• Navigate to rubrix/rubrix/web directory.
• Enter the following command in the terminal to launch web application:

  $ python app.py

3. Dockerized Web Application

This is for if you want to deploy rubrix on a server e.g. an Ubuntu Linux server on AWS

1. Navigate to the top directory
2. Enter the following command to build the docker image:

 $ sudo docker build -t <YOUR-NAME>/rubrix .

3. You can then run:

 $ sudo docker run -p 9000:80 <YOUR-NAME>/rubrix

The ideal setup for this would be to have a Apache/Nginx reverse proxy setup on the host system, pointing to port 9000 in this case, and the host system's Apache/Nginx would handle SSL. This would be so you can deploy the application over and over again without worrying about remaking SSL certificates.

The Dockerfile does not use the environment.yml file because using conda on any sort of production environment is a nightmare. Changes made there will not be reflected in the Dockerized container.

Contributing Guidelines

There are no specific guidelines for contributing, apart from a few general guidelines we tried to follow, such as:

• Code should follow PEP8 standards as closely as possible
• We use Google-Style docstrings to document the Python modules in this project.

If you see something that could be improved, send a pull request! We are always happy to look at improvements, to ensure that rubrix, as a project, is the best version of itself.

If you think something should be done differently (or is just-plain-broken), please create an issue.

License

See the LICENSE file for more details.