Image and Video Synthesis: How, Why and "What if"?

Generative modeling approaches are now at the point where high definition images can be synthesized from noise vectors and conditional methods enable video synthesis and future prediction. These technologies are reaching the point when they work well enough to both fascinate and disturb the general public, and to provide a rich unexplored medium of expression for artists.

While the end results seem similar, the approaches taken in visual synthesis range from conditional generative adversarial networks, through variational auto encoders to traditional graphics tricks of the trade. Moreover, the goals of synthesis research vary from modeling statistical distributions in machine learning, through realistic picture-perfect recreations of the world in graphics, and all the way to providing tools of artistic expression.

Additionally, there is a disconnect between research aimed at synthesis and practitioners interested in forensics. The issue of fake content synthesis and detection has recently become relevant to the public at large as a result of current political and social trends, and we can no longer afford to operate in two parallel universes.

1. To bring together experts from multiple disciplines working on and with synthesis to learn from each other.
2. To encourage discussion on the underlying goals of video and image synthesis, the ethics surrounding the resulting artifacts and the need for forensics to separate the fake from the real.

The program will consist of invited talks from research leaders and leading digital artist on the topics of synthesis and forensics, followed by a panel discussion on ethics and societal issues in synthesis algorithms. Speakers are encouraged to discuss ethical issues in their talks.

TBD

Shiry Ginosar, UC Berkeley

Taesung Park, UC Berkeley

Jun-Yan Zhu, MIT

Ming-Yu Liu, NVIDIA Corp.

Aaron Hertzmann, Adobe Inc.