This repository contains a project where Generative Adversarial Networks (GAN) and Variational Autoencoders (VAE) are trained on a custom dataset to generate images. The results from the two models are then compared to evaluate their performance in terms of image generation quality.
Objective:
- Train GAN and VAE models on a custom image dataset.
- Generate new images using both models.
- Compare the generated images to analyze the differences and strengths of each model.
Models Implemented:
- GAN (Generative Adversarial Network): A model composed of a Generator and a Discriminator, where the generator tries to create realistic images and the discriminator attempts to distinguish between real and generated images.
- VAE (Variational Autoencoder): A type of autoencoder that learns to generate images by mapping input data into a latent space and reconstructing from that space.
| Criteria | GAN | VAE |
|---|---|---|
| Image Sharpness | High sharpness, realistic details | Lower sharpness, often blurry |
| Training Stability | Unstable, requires careful tuning | Stable and predictable |
| Latent Space | Less structured, less interpretable | Well-organized, smooth interpolation |
| Image Diversity | Risk of mode collapse, limited diversity | High diversity, good variation |
| Training Time | Longer, due to unstable convergence | Faster and stable |