GroceryNet is a web-based application for grocery product classification and segmentation using a dual-input parallel CNN model. It leverages multicolor spaces and pre-trained ResNeXt50_32x4d encoders to achieve accurate results. This project was developed as part of a thesis titled "Grocery Product Image Analysis using Multicolor Dual-Input Parallel CNN for Simultaneous Semantic Segmentation and Classification".

1. Overview
2. Features
3. Installation
4. Usage
5. Dataset and Model
6. Download Links

GroceryNet is designed to classify and segment grocery product images using a deep learning model. The application provides:

• Image Classification: Predicts the class of a grocery product.
• Semantic Segmentation: Highlights the product in the image.
• Top-5 Predictions: Displays the top 5 probable classes with confidence scores.

The model is built using PyTorch and Streamlit for the web interface.

• Dual-Input CNN: Utilizes RGB and XYZ color spaces for improved accuracy.
• Pre-Trained Encoders: Uses ResNeXt50_32x4d encoders for feature extraction.
• Interactive Web Interface: Built with Streamlit for easy user interaction.
• Top-5 Predictions: Visualizes the top 5 predictions with confidence scores.
• Segmentation Overlay: Displays the segmentation mask over the input image.

• Python 3.8 or higher (https://www.python.org/downloads/)
• pip (Python Package Manager)
• Jupyter Notebook (optional, for going through how the model was trained)
• Git (optional, for cloning the repository)
• Streamlit (https://streamlit.io/)
• A code editor (e.g., VSCode or PyCharm)

1. Clone the Repository (Optional): If you have not already downloaded the project files, you can clone the repository, where the “your-username” is your github username:

   git clone https://github.com/your-username/GroceryNet.git 
   cd GroceryNet
   

Alternatively, if the files were manually downloaded, ensure you are in the correct project directory.

2. Set Up a Virtual Environment (Optional but Recommended): To avoid package conflicts, create and activate a virtual environment:

   Windows:

    python -m venv venv          # for Mac/Linux: python3 -m venv venv
    venv\Scripts\activate        # source venv/bin/activate
   
   

3. Download the Model and Dataset:

• Download the pre-trained model (best_run.pth) and dataset from the Google Drive link.

• Place the best_run.pth file in the root directory of the project.

• Extract the dataset and place it in the appropriate directory (e.g., Hierarchical Grocery Store Dataset with Masks).

• Run the Application: streamlit run "app.py"

Example:

• Upload an Image: Use the file uploader in the web interface to upload a grocery product image (JPG or PNG format).

The application will display:

• The input image.

• The segmentation map overlaid on the image.

• The top-5 predictions with confidence scores.

Dataset: The dataset used for training and testing is the Hierarchical Grocery Store Dataset (Klasson et al., 2019). It contains images of grocery products along with their corresponding masks.

Model: The model is a Parallel U-Net with pre-trained ResNeXt50_32x4d encoders. It was trained on the dataset mentioned above and achieves high accuracy in both classification and segmentation tasks.

• Model: Download "best_run.pth" https://drive.google.com/file/d/1Ew-54l0w0p4EDOTzy-urv_eaVSQ-wP2s/view?usp=sharing

• Dataset: Download and extract "Hierarchical Grocery Store Dataset with Masks.zip" https://drive.google.com/file/d/1swV1AqYNt6miggjxV9pImcS4W_Ef8DB9/view?usp=sharing

• Model Training: Download "GroceryNetImplementation.ipynb" https://drive.google.com/file/d/1w1wdFu67Zuy9zPtBJrKUnmYyXjSt-mjN/view?usp=sharing

Note: The "Model Training" link is provided for those who want to go through how the model was trained.

University of the Philippines Tacloban College: For providing the resources and support for this project.

• Streamlit: For the easy-to-use web framework.

• PyTorch: For the deep learning framework.

• Segmentation Models PyTorch: For the U-Net implementation.