This repository offers a collection of notebooks implementing various Variational AutoEncoder (VAE) architectures using the Keras deep learning framework. These implementations span multiple datasets and offer insights into the capabilities of VAEs for tasks such as neural compression and generative modeling.
This notebook provides a complete pipeline for building, training, and evaluating a 0 to 9), each having dimensions of 28x28 pixels.
In this notebook, you'll find the necessary code for creating, training, and assessing a 32x32 pixels, distributed across 10 different classes, each with 6000 images. The primary goal is to employ VAEs for encoding these images into a lower-dimensional latent space and then reconstructing them. The notebook also evaluates the model by comparing the original and reconstructed images.
This notebook focuses on the application of 96x96 pixel images across 10 classes. Unlike other similar datasets, STL-10 offers both labeled and unlabeled data, making it ideal for exploring feature learning techniques on higher-resolution images.
For a more detailed understanding, each notebook includes extensive commentary and evaluation metrics to gauge the model's performance. Feel free to explore!