I have added my implementation of 2 popular generative models GAN and VAE, I trained and comapred different configuraitons of these models.
GAN (Generative Adversarial Networks) and VAE (Variational Autoencoder) are two popular generative models which are currently being used widely in applications like deepfakes, music generation. They both have their own merits and demerits. I performed extensive experimentation on both generative models, tried different type of layers for both, tried different activation function, different latent variable size.
The notebooks were created on colab, you can either upload them on colab or run on any other system. The code is cuda enabled, it'll switch to cuda if there's one available. Please contact me if you face any problem while running the code.
VAE best result:
GAN best result:
