Develop your own Generative Adversarial Network - An introduction to GANs with TensorFlow Keras

About this event

Generative Adversarial Networks (GANs) were first introduced by Ian Goodfellow and his colleagues in June 2014,  where two neural networks contest with each other in a game. With the invention of GANs, Generative Models had started showing promising results in generating realistic images. GANs has shown tremendous success in Computer Vision. 

In this short talk, I will describe the basic idea behind adversarial learning in its most basic form. After that I will show live some code and how easy is to implement with Keras GANs. I will also point out which Keras features make the implementation easy and point out what resources are available for learning more about this topic. 

*No previous Keras experience is required (although it would be very helpful) as I will try to highlight the main components of the GANs in the Keras code to make it understandable also for participants not familiar with TensorFlow.

Event Wrap-up:

You can find the recording of the event on our YouTube channel and the slides and code examples on GitHub. As promised, here are some additional resources that Umberto would like to share with you:

  1. GANs for coloring images: Here is a very nice tutorial that you can run and paper with explanations.
  2. GANs for superresolution: Explanation resources one and two. Papers one and two.
  3. Neural Style Transfer that he showed in the talk.
  4. GitHub Repository of the two days lecture he gave in London at the AI O'Reilly Conference on CNNs.

Speaker

When

Thursday, Jan 27
6:00 PM - 7:00 PM (CET)

Organizers

Partners