Semantic Image Search

Overview

semantic-image-search is a web-based application that allows users to perform semantic searches on a collection of images using either textual descriptions or image inputs. It leverages the OpenAI's CLIP model for embedding extraction and Milvus for efficient similarity search.

Features

• Collection Management: Create, upload, and delete image collections.
• Search Capabilities: Perform image and text-based searches on image collections.
• Efficient Storage and Retrieval: Utilize Milvus for fast and scalable vector similarity search.

Installation

Prerequisites

• Python 3.8+
• pytorch
• streamlit
• transformers
• pymilvus

Setup

1. Clone the repository:

   git clone https://github.com/sefaburakokcu/semantic-image-search.git
   cd semantic-image-search

2. Install the required packages:

   pip install -r requirements.txt

3. Run the application:

   streamlit run app.py  --server.maxUploadSize 2000

Usage

Sidebar: Collection Management

• Create Collection: Enter a name for the new collection and click "Create Collection."
• Upload Images to Collection: Select an existing collection, upload images or a zip file containing images, and click "Upload Images."
• Delete Collection: Select a collection to delete and click "Delete Collection."

Main Interface: Image and Text Search

• Search Configuration:
  • Select the collection to search.
  • Set the number of results and the number of results to display per row.
  • Choose between text search and image search.
• Search Input:
  • For text search, enter a search query.
  • For image search, upload an image file.
• Search Results: Click "Search" to view results. Images are displayed with their similarity percentage.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

• Streamlit for the web application framework
• Milvus for the vector database
• OpenAI for the CLIP model from HuggingFace transformers

For any questions or issues, please open an issue on GitHub.