Image Classification

Description

This project is an image classification application using Convolutional Neural Networks (CNNs). It is a learning platform for me to get deep hands-on experience working with production-ready AI platforms. Check the TODO section for what's planned.

Project Structure

• data/: Contains datasets and images.
• src/: Source code for data preprocessing, model building, and prediction.
• results/: Stores image output, models, and logs.

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/ImageClassificationProject.git
   cd ImageClassificationProject

2. Install dependencies:

   pip install -r requirements.txt

Usage

• Train the model:

  python -m src.train

  This will create two models - one using data augmentation, and one without.

  Note: If the model(s) already exist(s), delete/rename them from /results/model/ or run

  python -m src.train -f

• Compare the models:

  python -m src.compare

  This runs an evaluation of the two models and reports Loss and Accuracy:

  Model without Data Augmentation: Loss = 0.9609, Accuracy = 0.6886
  Model with Data Augmentation:    Loss = 0.8786, Accuracy = 0.7005
  

  The program will then run a prediction on a number of test images using both models. The test images can be found in /results/output/<Ymd-HMS>/

  Note: comparison can be run during the above training step by using python -m src.train -c

• Predict an image:

  from src.predict import predict, get_class_probabilities, print_class_probabilities
  
  image_path = 'path_to_your_image.jpg'
  predictions = predict(image_path)
  class_probabilities = get_class_probabilities(predictions, class_names)
  print_class_probabilities(class_probabilities)

TODO

• Model Improvement:

  • Apply data augmentation techniques
  • Implement regularization techniques (dropout, batch normalization)
  • Perform hyperparameter tuning

• Advanced Model Architectures:

  • Use transfer learning with pre-trained models (e.g., VGG16, ResNet50)
  • Experiment with deeper networks

• Expanding the Dataset:

  • Create and label a custom dataset
  • Fine-tune the model on specific classes

• Real-Time Object Classification:

  • Integrate the model with a webcam for real-time classification
  • Deploy the model on a mobile device

• Visualization and Interpretation:

  • Visualize feature maps from convolutional layers
  • Generate Class Activation Maps (CAMs)

• Evaluation and Reporting:

  • Create a confusion matrix
  • Calculate detailed metrics (precision, recall, F1-score)

• Application Development:

  • Implement a CLI to run the project
  • Develop a web application for image upload and classification
  • Create an API for the image classification model

Contributing

Contributions are welcome! Please submit a pull request.

License

This project is licensed under the MIT License.