GANs for MNIST Digit Generation

2020

Diagram of the generative adversarial network pipeline with generator creating fake digits from latent noise and discriminator classifying real versus generated MNIST samples

Architecture diagram of the generator with Leaky ReLU and tanh layers, and discriminator with Leaky ReLU and sigmoid output

A 4x4 grid of synthetic handwritten MNIST digits produced by the trained generator network

Project information

Project URL: GitHub Repository

About this project

A Generative Adversarial Network (GAN) implemented in PyTorch trained on the MNIST handwritten digits dataset. The generator network synthesizes new, realistic 28x28 grayscale digits from random latent noise vectors, while the discriminator is trained to distinguish authentic MNIST samples from generated images.

Back to Projects