Digit Generation using Wasserstein GAN (W-GAN)

• Concepts used:
  • Generator
  • Critic (Discriminator/Classifier)
  • Batch Normalization
  • Transpose Convolution
  • Gradient Penalty (To prevent mode collapse)
  • Wasserstein loss
• Tech Stack:
• Python and PyTorch
• GitHub: Project Link

Trained the model on MNIST dataset using Wasserstein Generative Adversarial Neural Network(GAN) in Python using PyTorch framework and our model was able to produce great results than the previous state of the art GAN's. It tackle the issue of mode collape(when the generator can only produce a single type of output or a small set of outputs) and now we can train our model for more epochs without the facing the issue of Mode Collapse and model can generate better results than the previous GAN's.