This repository contains implementations of Variational Autoencoders (VAE) and their application in image classification tasks, primarily focusing on the Fashion MNIST dataset.
- Implements Variational Autoencoders (VAE) for generating and reconstructing images.
- Utilizes TensorFlow and Keras for building and training models.
- Supports dimensionality reduction for improving image classification using K-Nearest Neighbors (KNN).
- Includes detailed performance evaluation with confusion matrices and classification reports.
- Provides visualizations of training losses, latent spaces, and generated images.
- Clone the repository.
- Install the necessary dependencies using
pip install -r requirements.txt. - Ensure TensorFlow with GPU support is installed if GPU processing is desired.
The primary dataset used is Fashion MNIST, which includes 60,000 training images and 10,000 testing images of 10 fashion categories.
- Execute the VAE training script to learn latent representations of images.
- The model automatically performs image reconstruction and generation.
- Use the encoded representations from VAE as features for training a KNN classifier.
- Evaluate the classifier's performance using the test dataset and calculate various metrics like accuracy, precision, recall, and F1-score.
- Check the output directory for training logs, model checkpoints, and generated images.
- Review the classification reports and confusion matrices to understand model performance.
Contributions, issues, and feature requests are welcome! Feel free to check the issues page.
This project is available under the MIT License. See the LICENSE file for more details.
- TensorFlow and Keras documentation for providing extensive guides and API documentation.
- Fashion MNIST dataset creators for providing a benchmark dataset for image classification tasks.
For more details, please visit the GitHub repository.