Deepfake Detection: Humans vs. Machines
COM1 Level 2
SR2, COM1-02-04 and via zoom (Hybrid)
Abstract:
Practically anyone can now generate a realistic looking deepfake video. Many methods to detect deepfakes were recently proposed by the research community. However, it is still unclear how realistic deepfake videos are for an average person and whether the algorithms are significantly better than humans at detecting them. In this talk, we will compare the results of a subjective study (60 naive subjects) and several deepfake detection algorithms in terms of the human and machine ability to distinguish real videos from different types of deepfakes. We will also discuss the generalization, attribution, and interpretability problems faced by the algorithms detecting deepfakes of unknown origin.
Biodata:
Pavel Korshunov received his PhD from School of Computing, National University of Singapore in 2011, completed several years of postdoc at EPFL (Switzerland), and since 2015 is a research associate at Idiap Research Institute (Switzerland). His current interests include deepfake detections, audio-visual age verification, and image morphing. Previously, he also worked on video tampering detection, visual privacy protection, high dynamic range imaging, speech anti-spoofing, and crowdsourcing. He received an ACM TOMM journal best paper award (2011), two top 10% best paper awards in MMSP 2014, and a top 10% best paper award in ICIP 2014. He has over 70 research publications and is a co-editor of JPEG XT standard for HDR images.