Package-Recognizer

A deep learning-based image classification application that recognizes and categorizes daily packaging items. Built with FastAI and deployed using Gradio on Hugging Face Spaces.

Table of Contents

• Overview
• Features
• Models Used
• Gradio Interface
• Supported Categories
• Quick Start
• Project Structure
• Development
• Data & Models
• Technologies Used
• License
• Contributing
• Contact & Support

Overview

Package-Recognizer is a computer vision project designed to automatically classify packaging images into 12 distinct product categories. This application can be used for waste management, recycling sorting, inventory management, and product recognition tasks.

Features

• Multi-class Image Classification: Recognizes 12 packaging categories
• Fast & Accurate: Powered by FastAI pre-trained models
• Easy-to-Use Interface: Interactive Gradio web interface
• Cloud Deployment: Hosted on Hugging Face Spaces for easy access
• Jupyter Notebooks: Complete workflow from data preparation to inference
• Reproducible Pipeline: Well-documented data cleaning and training process

Models Used

• ResNet34: Primary model for deployment
• MobileNetV3_Large_100: Alternative model for experimentation

Gradio Interface

Upload an image of a packaged product and receive the predicted category along with confidence scores.

Categories

1. Bottled Water
2. Soft Drink Cans
3. Milk Cartons
4. Snack Chips Packets
5. Chocolate Bars
6. Instant Noodles Packets
7. Toothpaste Boxes
8. Shampoo Bottles
9. Soap Bars
10. Cooking Oil Bottles
11. Coffee Jars
12. Detergent Packs

Quick Start

Try the Live Demo

Visit the deployed application on Hugging Face Spaces: Package-Recognizer on HF Spaces

Local Installation

Prerequisites

• Python 3.12.11 or higher
• pip or conda

Setup

1. Clone the repository:

git clone https://github.com/ashfaqfardin/Package-Recognizer
cd Package-Recognizer

2. Install dependencies:

pip install -r deployment/requirements.txt

3. Download the pre-trained model from Google Drive:

   • Models: https://drive.google.com/drive/folders/1OxZww1lAyfiMygizYQcjqISgzzx_Ydl2?usp=sharing
   • Place pkg_recognizer_v1.pkl in the deployment/ directory

4. Run the application:

cd deployment
python app.py

The application will launch at http://localhost:7860

Project Structure

Package-Recognizer/
├── deployment/              # Production application
│   ├── app.py              # Gradio interface
│   ├── app.ipynb           # Jupyter version
│   ├── requirements.txt     # Dependencies
│   └── README.md
├── notebooks/              # Development workflow
│   ├── data_prep.ipynb     # Data preparation
│   ├── training_and_data_cleaning.ipynb  # Model training & data cleaning
│   └── inference.ipynb     # Inference examples
├── models/                 # Pre-trained models
│   └── README.md          # Model documentation & download links
├── data/                  # Training datasets
│   └── README.md         # Data documentation & download links
├── docs/                 # Documentation
│   ├── index.md
│   ├── pkg_recognizer.html
│   └── _config.yml
└── README.md            # This file

Development

Workflow

1. Data Preparation (notebooks/data_prep.ipynb):

   • Load and explore datasets
   • Prepare data for training

2. Training & Data Cleaning (notebooks/training_and_data_cleaning.ipynb):

   • Clean and preprocess data
   • Train the FastAI model
   • Export model to .pkl format

3. Inference (notebooks/inference.ipynb):

   • Test the trained model
   • Generate predictions

Running Notebooks

All notebooks are Jupyter-based and can be run locally:

jupyter notebook notebooks/

Data & Models

Data

• Download extracted datasets: https://drive.google.com/drive/folders/1qeTjO90zWc45lGwgF7DqgLb-mmiFoLj-?usp=sharing
• Place in the data/ directory before running training workflows

Pre-trained Models

• Download models: https://drive.google.com/drive/folders/1OxZww1lAyfiMygizYQcjqISgzzx_Ydl2?usp=sharing
• Primary model: pkg_recognizer_v1.pkl (used for deployment)
• Place in the deployment/ directory for app.py or relevant notebooks

Technologies Used

• FastAI (2.8.4): Deep learning library
• Gradio (5.49.1): Web interface framework
• Python (3.12.11): Programming language
• Jupyter: Interactive notebooks

License

MIT License - See LICENSE file for details

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Contact & Support

For questions or issues, please open an issue on the repository. Email: imashfaqfardin@gmail.com