Deep Convolutional Generative Adversarial Networks
Project description:
Input: The CIFAR10 dataset.
Generator: It takes a random vector from the latent space and map it to a synthesized image.
Discriminator: It maps the image to a binary score that is the probability of being fake/real image.
Goal: The model is trained by generating fake images produced by generator, mixing them with real images and adding a small random noise to their corresponding labels. Generator should be able to fool the discriminator.
The grid of decoded images are shown as follows which are difficult to distinguish:
The source code of this project is available in my Github page.