Retail Product Recognition System

1. Dataset Sources

• Coke Classification Dataset
• Soda Project Dataset
• Snacks Dataset
• Dabur Active Juice Dataset

Dataset Structure

Original Dataset

New Dataset

2. Model Information

A ResNet50 model is utilized to identify products across 15 retail products within 3 categories. This model processes the images provided by users through the frontend.

Model Training Metrics

Below are the visualized examples of the model's training metrics:

Precision - 98.08%

Recall - 98.02%

F1 Score - 98.01%

Train vs. Validation Loss

3. Frontend

The frontend is developed using ReactJS, providing functionality to upload an image and receive the top 3 predictions for that image. It includes the product category, exact product name, and classification confidence.

4. Backend

The backend is developed with Python and FastAPI. It initializes the model and exposes the prediction API to the frontend. When a user uploads an image, the frontend relays it to the backend via FastAPI, which handles product classification. Finally, the backend forwards the predictions back to the frontend.

5. API Reference

1. Frontend: http://<machine_ip>:3000
2. Backend: http://<machine_ip>:8000
3. Prediction Endpoint: http://<machine_ip>:3000/predict/ (POST) - Used through the frontend

6. Instructions to Run the Retail Product Recognition Application

1. Ensure you are using an x86_64 machine.
2. Make sure you have Docker and Docker Compose installed.
3. Clone the repository with the following command:

   git clone <repository_url>
   cd <repository_name>

4. Build and run the application using Docker Compose:

   docker compose build
   docker compose up -d
   

5. Open your browser:
   • Frontend (React): http://<machine_ip>:3000 - (IMPORTANT: Use CORS blocker extension for your browser)
   • Backend (FastAPI): http://<machine_ip>:8000
6. Stopping the Project. To stop the project, run:

   docker-compose down