Reverse Image Search Engine(Local)

This is a simple reverse image search tool that uses the Streamlit framework. It allows users to find similar images locally.

Why is it helpful?

This tool can help identify similar images. This uses EfficientNet-B0 for vector embedding and Chromadb(vector database) for vector storage and retrieval. EfficientNet-B0 is a lightweight convolutional neural network architecture that is designed to be efficient and accurate for image recognition tasks.

Installation

To use the reverse image search engine, simply follow these steps:

# clone the repo
git clone https://github.com/tikendraw/reverse-image-search.git

# go inside
cd reverse-image-search

# install with pip
pip install . 

Usage

1. Via browser bash command

   # bash command 
   image_search

   or just run the launch.py file

   python launch.py

2. Via command line

• Embed a directory

  image_search_cli embed create -d <directory_path>

• Search for similar images

  image_search_cli search <image_path> -n <num_similar_images>

• Search for multiple images

  image_search_cli search <image_path1> <image_path2> <image_path3> ...

• Update embeddings for a directory

  image_search_cli embed update -d <directory_path>

• Delete embeddings for a directory

  image_search_cli embed delete -d <directory_path>

  Use delete_all_embeddings string to delete all at once

• Delete all embeddings

  image_search_cli embed delete -d delete_all_embeddings

Screenshot

1. Web

2. CLI

Tech stack use

The reverse image search engine is built using the following technologies:

• Python
• transformers
• pytorch
• EfficientNet-B0 for vector embedding
• Chromadb(vector database)
• Streamlit
• Pillow

How to contribute

Future updates:

• Facial Recognition
• Search Images with words

If you would like to contribute to the reverse image search engine, please feel free to open a pull request.

Citation

1. EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

@misc{tan2020efficientnetrethinkingmodelscaling,
      title={EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks}, 
      author={Mingxing Tan and Quoc V. Le},
      year={2020},
      eprint={1905.11946},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1905.11946}, 
}

License

The reverse image search engine is licensed under the MIT License.

I hope this is helpful!