This repository presents a convolutional implementation of the Variational Deep Embedding: An Unsupervised and Generative Approach to Clustering paper. The primary goal of this project is to provide a straightforward and well-commented code base that serves as an entry point for understanding the VaDE architecture and its mathematical underpinnings.

The implementation is tailored for the clustering of landscape images from a Kaggle dataset.

The core purpose of this repository is to offer a compact and comprehensible codebase that helps users grasp the inner workings of the VaDE model. It's designed to help the comprehension of the original work's repository by providing a compact and well commented, albeit specific, example.

⚠️ Note: This is not a "ready-to-use" script. The code is highly specific to my project's task, and there is currently no configuration file for easy customization. The network structure is hard-coded to solve my personal task. I may, however, improve this aspect in the future.

• train_VaDE.py: The main Python script containing the implementation of the convolutional VaDE model. This is where you'll find the architecture and training loop.
• VaDe_evaluate.ipynb: A Jupyter Notebook that showcases the results of the specific clustering task. While you're free to ignore it, it offers a peek into the model's performance on the landscape image dataset. (As noted, the results are decent but could be significantly improved with better computational resources.)

This project was developed for an exam of my Master's Degree in Applied Physics at the University of Bologna (UNIBO).