CelebA-GAN

In this project, I developed a GAN to generate realistic face images using the CelebA dataset. The model comprises a Generator and Discriminator, with the generator synthesizing images from random noise and the discriminator distinguishing real images from generated ones. The CelebA images were preprocessed, and both models were trained using Binary Cross-Entropy Loss with Adam Optimizer for stable convergence.

The training loop updates the discriminator on both real and fake images to improve its classification, followed by training the generator to maximize its chances of "fooling" the discriminator. Progress images were saved periodically, allowing visual tracking of the generator’s improvements over time. This project provided a hands-on understanding of GANs, including adversarial training dynamics and model optimization for realistic image generation on resource-constrained devices.