MLP or Vanila GAN & DCGAN
Trained with NVIDIA P100 GPU
- Used only Linear Layers (MLP) to create MNIST
- Paper Link by Ian J. Goodfellow
- Used Convolution and Tranposed Convolution
- Paper by Alec Radford, Luke Metz, Soumith Chintala (Link)
- From the beginning, the noises are just noises but after few epochs, it starts to learn how to fake the discriminator.
- Unlike Valina GAN, DCGAN produces more clear and less noise output at the end.
GAN was devloped only five years ago. There are numerous GANs so I will keep study them until I get the new idea.