GANs for Real-Life Video Applications – talk by Oran Gafni from Facebook AI Research

Abstract: This talk will focus on two research papers that demonstrate the use of generative adversarial networks for real-life video applications, through face and full-body synthesis. The first paper (“Live Face De-Identification in Video”), proposes a novel method for face de-identification, that enables fully automatic video modification at high frame rates, by maximally decorrelate the identity, while having the perception (pose, illumination and expression) fixed. The second paper (“Vid2Game: Controllable Characters Extracted from Real-World Videos”), extracts a controllable model from a given video of a person performing a certain activity. The model generates novel image sequences of that person, according to arbitrary user-defined control signals, typically marking the displacement of the moving body. The generated video can have an arbitrary background, and effectively capture both the dynamics and appearance of the person