Project 5: GANs

Overview

Training Generative Adversarial Networks (GANs) for image generation. This project covers the adversarial training paradigm with generator and discriminator networks competing to produce realistic images.

Vanilla GAN DCGAN Generator Discriminator Training Images

Notebooks

Notebook	Topic	Description
052	Vanilla GAN	Basic GAN architecture, adversarial training, image generation

Available Models

Model	File	Description
Generator	`generator_99.pth`	Trained generator weights (epoch 99)
Discriminator	`discriminator_99.pth`	Trained discriminator weights (epoch 99)

Key Techniques

Adversarial Training - Minimax game between generator and discriminator
Generator Network - Transforms latent vectors to images via transposed convolutions
Discriminator Network - Binary classifier distinguishing real from fake images
Training Stability - Techniques for stable GAN training (batch norm, LeakyReLU)

Available Tools

Tool	Description	Variation
`load_proj5_info`	Get Project 5 (GANs) information and resources	V3
`list_proj5_models`	List GAN models (generator/discriminator)	V3