This repository contains a simple variational autoencoder used to generate handwritten digits based on the MNIST dataset.
The architecture of a simple CNN-based VAE has been modified according to the idea of conditioning, based on the paper Learning Structured Output Representations from Attributes using Deep Conditional Generative Models. This allows generating arbitrary digits by providing class labels to the decoder.
Important
The project is managed by uv. I strongly recommend using it for installation and usage, not only for this project but for all your Python software.
To run the notebook:
- Clone the repository:
git clone https://github.com/integraledelebesgue/conditional-vae.git
cd conditional-vae- Install the dependencies and create the virtual environment:
uv sync