Visual Search for E-Commerce Store 🛍️

Overview 🔍

This project enables visual search for an e-commerce store, where users can upload an image and receive similar product recommendations. Instead of typing a text query, the user simply uploads an image (e.g., a pair of swimming trunks 🩳), and the system will analyze the image and return matching products from the catalog.

Technologies Used ⚙️

1. Frontend:

   • Tools: HTML, CSS, JavaScript
   • Function: The user interface where users upload images. The frontend sends the image to the backend for processing and displays the results.

2. Backend:

   • Tool: Flask (Python)
   • Function: Handles incoming requests, processes the uploaded image, interacts with Pinecone to search for similar products, and sends results back to the frontend.

3. Image Embedding:

   • Tool: ResNet50 (Pre-trained deep learning model)
   • Function: Converts the uploaded image into an embedding (a vector of numbers) that captures its unique features, making it comparable with other product images.

4. Embedding Storage & Similarity Search:

   • Tool: Pinecone
   • Function: Stores embeddings for all catalog items, enabling fast similarity searches.
   • Search Criteria: The system retrieves the top 5 similar items based on cosine similarity, with a threshold score of 0.70 to ensure relevance.

5. Metadata Storage:

   • Tool: SQLite
   • Function: Stores metadata for each product (e.g., ID, name, price). After retrieving similar embeddings from Pinecone, Flask uses the product IDs to fetch complete details from SQLite.

6. Result Display:

   • Function: The backend sends the product details (name, price, etc.) and images back to the frontend, which displays the top-matching products.

Project Structure 📁

Visual Search/
   ├── app.py               # Main Flask backend file
   ├── templates/
   │   └── index.html       # Frontend HTML file
   ├── static/
   │   ├── Images/          # Folder containing product images
   │   ├── script.js        # JavaScript for frontend functionality
   │   └── style.css        # CSS for frontend styling
   ├── catalogs/            # Product catalog images and metadata
   ├── Product.db           # SQLite database with product metadata
   ├── .env                 # Environment file with Pinecone API key
   └── requirements.txt     # List of required Python packages

Getting Started 🚀

Prerequisites 🛠️

1. Python (>= 3.6)
2. Flask (Web framework)
3. ResNet50 (Pre-trained deep learning model)
4. Pinecone (Vector database for similarity search)
5. SQLite (Relational database for metadata)

Setup 🧑‍💻

1. Clone the repository to your local machine.

2. Install the required dependencies:

   pip install -r requirements.txt

3. Create a .env file in the root directory and add your Pinecone API Key:

   PINECONE_API_KEY=your_pinecone_api_key
   

Running the Application 🚂

1. Start the Flask app:

   python app.py

2. Open your browser and go to http://127.0.0.1:5000 to interact with the visual search.

---

Demo

visual-search_Dcp9ejUz.mp4