Frank or Not Frank

This project uses a deep learning model to identify if an image contains Frank the dog or not. The model is based on the VGG16 architecture and uses transfer learning for binary classification.

Table of Contents

• Installation
• Usage
• Training
• Inference
• Contributing
• License

Installation

Prerequisites

• Python 3.7+
• PyTorch
• torchvision
• OpenCV
• Pillow

Clone the Repository

git clone https://github.com/jbfly/frank_or_not.git
cd frank_or_not

Install Dependencies

You can install the required dependencies using pip:

pip install torch torchvision opencv-python pillow

Usage

Inference

To run the model on your local machine using your webcam, make sure you have the model file best_frank_or_not_model_run_1.pth in the same directory as frank_or_not.py. Then execute:

python frank_or_not.py

This will open a window showing the webcam feed with the prediction overlaid on the video.

Training

If you need to retrain the model, you can use the train_and_evaluate.py script. The training script is set up to handle multiple training runs and implements early stopping.

Ensure your dataset is organized as follows:

dataset/
    train/
        frank/
        not_frank/
    valid/
        frank/
        not_frank/

Then, execute:

python train_and_evaluate.py

This script will perform multiple training runs, save the best model based on validation accuracy, and store it as best_frank_or_not_model_run_X.pth.

Inference

The frank_or_not.py script performs real-time inference using your webcam. Make sure the model file is in the same directory as the script and run:

python frank_or_not.py

Example Output

The script will display a window with the webcam feed and overlay the prediction ("Frank" or "Not Frank") on each frame.

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/fooBar)
3. Commit your changes (git commit -am 'Add some fooBar')
4. Push to the branch (git push origin feature/fooBar)
5. Create a new Pull Request

License

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