ML-Project-GAN

Repository for the Machine Learning Project, Spring 2018.

Project Introduction Presentation

Progress Presentation

Final Presentation

Team

Lydia, Kevin, Adriana

GAN, Tensorflow and MNIST

To start, we reproduced the work done by Wiseodd, specifically their vanilla_gan in Tensorflow.

Goals

• Flowers Dataset
  • Create fake flower images of the various types of flowers
• Faces Dataset
  • Create faces of the various emotions.
  • Advanced: Use a specific starter face and manipulate it into the various emotions
• GANs are hard – what if we fail?
  • Forget about the GAN and use a CNN to predict facial emotions

Code Modifications

• Change the structure of the tensors to match the dataset
• Change the 2-layer neural net to a CNN
• Advanced: Face manipulation
  • Use starter images instead of noise