This system is designed to perform semantic image retrieval based on textual queries. Utilizing the CLIP (Contrastive Language-Image Pretraining) model, it generates embeddings for text queries and images within a dataset, allowing for effective and efficient similarity searches.
- Description: Downloading the images and a first look.
- Description: Manages the generation of embeddings for the dataset images using the CLIP model.
- Functionality: Encodes images into embeddings which are stored for later search comparisons.
- Description: Responsible for uploading the generated embeddings and image metadata into the Qdrant vector database.
- Functionality: Ensures embeddings are stored in Qdrant, ready for similarity searches.
- Description: Acts as the central API endpoint for the system.
- Functionality: Handles user text queries, performs vector searches in Qdrant, and returns the most relevant image results.
- Description: Provides a FastAPI server setup to serve the Text2Image search functionality over a REST API.
- Functionality: Interfaces with
main.pyto accept queries and return image results through HTTP requests.
A minimalistic frontend that allows users to enter text queries and displays relevant image results fetched from the backend.
Clone this repository to your desired location:
git clone https://github.com/Oguzhan1996/Text2Image-Search.git
Navigate to the project directory and install the required Python packages:
cd Text2Image-Search
pip install -r requirements.txt
run Exploratory Data Analysis.ipynb to download the images
Run docker
Ensure Qdrant is running. If not, start your Qdrant instance as per the official documentation.
Running the Application
Generate and Upload Image Embeddings:
python embed_images.py
python upload_qdrant.py
Start the flask server by running the file FLASK.py
navigate in your browser to:http://127.0.0.1:5000
Here you can search for queries with the search bar