This project implements a Deep Convolutional Generative Adversarial Network (DCGAN) using PyTorch to generate realistic face images, trained on the CelebA dataset. The model is trained on the CelebA dataset and demonstrates the fundamental principles of adversarial training between a Generator and a Discriminator.
- Trains a DCGAN model with custom Generator and Discriminator networks.
- Initializes weights as per the DCGAN paper (Normal(0, 0.02)).
- Uses Binary Cross Entropy (BCE) loss for adversarial training.
- Visualizes intermediate samples to track model progression.
- Saves generated images and training samples for post-analysis.