The MeVer DeepFake Detection Service: Lessons Learnt from Developing and Deploying in the Wild

• DOI: 10.48550/arxiv.2204.12816
• Publication Date: 2022-01-01
• Authors: Spyridon Baxevanakis; Giorgos Kordopati...; Panagiotis Galopo...; Lazaros Apostolidis; Killian Levacher; Ipek B. Schlicht; Denis Teyssou; Ioannis Kompatsiaris; Symeon Papadopoulos
• Keywords: Mainstream

Enabled by recent improvements in generation methodologies, DeepFakes have become mainstream due to their increasingly better visual quality, the increase easy-to-use tools and rapid dissemination through social media. This fact poses a severe threat our societies with potential erode cohesion influence democracies. To mitigate threat, numerous DeepFake detection schemes been introduced literature but very few provide web service that can be used wild. In this paper, we introduce MeVer service, detecting deep learning manipulations images video. We present design implementation of proposed processing pipeline involves model ensemble scheme, endow card for transparency. Experimental results show performs robustly on three benchmark datasets while being vulnerable Adversarial Attacks. Finally, outline experience lessons learned when deploying research system into production hopes it will useful other academic industry teams.