Generative Adversarial Networks and the Next Generation in Cat Videos

A collaboration of Adobe Research, University of California at Berkeley, Carnegie Mellon University, and MIT’s CSAIL has produced a system called “Cohere”, described in the presentation “GANgealing: GAN-Supervised Dense Visual Alignment”, which uses the machine learning technique of generative adversarial networks to automatically align overlays on moving features in videos.

This will have applications for image and video processing, augmented reality, and the production of deep fake videos. If you can immediately think of your own applications, you’ll find source code posted on GitHub and a ready-to-run “GANgealing Mixed Reality Playground” on Google Colab.

1 Like

The Eagles were on to something back when, we just need to substitute the word “trust” for “hide”. It looks like what used to be proof in the form of videos is about to go out the window. The implications are staggering (think criminal prosecution/reasonable doubt, for example).

You can’t hide your lyin’ eyes
And your smile is a thin disguise
I thought by now you’d realize
There ain’t no way to hide your lyin’ eyes
There ain’t no way to hide your lyin’ eyes
Honey, you can’t hide your lyin’ eyes

3 Likes