Education

Abstract: In this era of misinformation, deepfake video is the most realistic fake information of all. It is a technique that uses a pre-trained generative adversarial network (GAN) to automatically replace the face of one person in a video with the face of another person. It has become nearly imperceptible to the naked eye due to recent developments in hardware and software. It has now become a major source of concern for individuals all over the world, particularly on social networking sites like as Facebook, YouTube, Twitter etc. This recall for finding a better approach to detect deepfake videos. Over the years many convolutional approaches has been taken. Although there are only a few publically available dataset, those approaches were trained and tested on different datasets. Here we represent a comparative study on those various approaches. We begin by defining the term "Deepfake" and the motivation for researching this topic. Then we’ll go over the databases that are publicly available, as well as the database that we’re working with. Then we go over four convolutional techniques in details, including certain terms related to architectures. Following that, we showed related work on those architectures. Then, using the same dataset, we present a comparative study of those four convolutional techniques. Furthermore, we also discussed our long-term research strategy.