DCGAN with Wasserstein Loss and Gradient Penalty

This project implements a Deep Convolutional Generative Adversarial Network (DCGAN) enhanced with Wasserstein Loss and Gradient Penalty for stable training and improved generation quality.

File Descriptions

• train.py: Main training script, orchestrating the model training with dataset preparation and training loops.
• dataset.py: Handles dataset loading and preprocessing, tailored for the DCGAN model requirements.
• model.py: Defines the DCGAN architecture, including both Generator and Discriminator models.
• utils.py: Provides utility functions and classes to support model training and data manipulation.

References

• DCGAN Paper: "Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks"
• Wasserstein Loss Paper: "Wasserstein GAN"
• Gradient Penalty Paper: "Improved Training of Wasserstein GANs"