Face Generation with Variational Autoencoder (VAE)

Overview

This project uses a Variational Autoencoder (VAE) to generate human faces based on the CelebA dataset. A VAE is a generative model that learns to represent high-dimensional data (like images) in a lower-dimensional latent space, and then generates new data from this space.

Installation

To run this project, you need Python 3.8 and predominantly the following packages installed:

• torch
• torchvision
• matplotlib
• streamlit

Install the necessary packages using the following command:

pip install -r requirements.txt

Or the following:

pip install torch torchvision matplotlib streamlit

Dataset

The CelebA dataset is used, which contains over 200,000 celebrity images with 40 attribute annotations. Ensure the dataset is downloaded and placed in the /data/img_align_celeba/ directory for the script to access. Location of the dataset: https://drive.google.com/file/d/1rxGTimhTS34eiamhldhkoJrDoVmYps5P/view?usp=sharing

Files in the Project

• vae_model.py: Contains the VAE implementation.
• train_vae.py: Script to train the VAE on the CelebA dataset.
• app.py: Streamlit application to display generated images from the trained VAE model.

How to Run

1. Train the VAE model by running train_vae.py. This will save the model checkpoints and generate images in the runs/ directory.
2. Start the Streamlit application with streamlit run app.py. Load the trained model and generate new images through the web interface.

Usage

• Modify train_vae.py to change training parameters, model architecture, or dataset processing.
• Use app.py to visualize the generated images and interact with the trained model.