CNN Food Classifier

This repository contains code for training a Convolutional Neural Network (CNN) food classifier using Python. The classifier is trained using images from the Food-5K dataset and utilizes the VGG16 pre-trained network for feature extraction.

Requirements

• Python 3.x
• TensorFlow
• scikit-learn
• NumPy
• imutils

File Structure

• train.py: Script for training the logistic regression model using extracted features.
• build_dataset.py: Script for organizing the dataset into training, testing, and validation splits.
• extract_features.py: Script for extracting features from images using VGG16 and preparing data for training.
• config.py: Configuration file containing paths and constants used throughout the project.

Usage

1. Build Dataset: Run build_dataset.py to organize the dataset into the desired splits (training, testing, and validation).

python build_dataset.py

2. Extract Features: Run extract_features.py to extract features from images using VGG16 and save them in CSV format along with labels.

python extract_features.py

3. Train Model: Finally, run train.py to train the logistic regression model using the extracted features.

python train.py

Configuration

• config.py contains paths and constants used in the project. Adjust these paths if necessary.

Output

• Trained model will be saved as model.cpickle.
• Label encoder will be saved as le.cpickle.
• Extracted features will be saved as CSV files in the output directory.

Note

• Ensure that the Food-5K dataset is downloaded and placed in the appropriate directory (Food-5K by default).
• Modify the config.py file according to your dataset directory structure if different from the default.
• Additional configuration options such as batch size can be adjusted in config.py.

Credits

This project utilizes the Food-5K dataset and the VGG16 pre-trained network. Credits to the authors of these resources.

License

This project is licensed under the MIT License - see the LICENSE file for details.