generative-ai

variational_autoencoder.ipynb

• Implementation of VAE using PyTorch, trained on celebA faces dataset.
• Making images smile, using both original image space and latent space arithmetic.

gan_cifar10.ipynb

• Implementation of standard GAN on CIFAR-10 data.

gan_mnist.ipynb

• Implementation of standard GAN on MNIST data.
• Used the following to prevent discriminator from becoming too good for the generator to learn.
  • Simple architecture for discriminator
  • Low learning rate for the discriminator
  • Label smoothening
  • Updating generator more often