One Shot Face Swapping on Megapixels

Face swapping has both positive applications such as entertainment,human-computer interaction, etc., and negative applications such as DeepFakethreats to politics, economics, etc. Nevertheless, it is necessary tounderstand the scheme of advanced methods for high-quality face swapping andgenerate enough and representative face swapping images to train DeepFakedetection algorithms. This paper proposes the first Megapixel level method forone shot Face Swapping (or MegaFS for short). Firstly, MegaFS organizes facerepresentation hierarchically by the proposed Hierarchical Representation FaceEncoder (HieRFE) in an extended latent space to maintain more facial details,rather than compressed representation in previous face swapping methods.Secondly, a carefully designed Face Transfer Module (FTM) is proposed totransfer the identity from a source image to the target by a non-lineartrajectory without explicit feature disentanglement. Finally, the swapped facescan be synthesized by StyleGAN2 with the benefits of its training stability andpowerful generative capability. Each part of MegaFS can be trained separatelyso the requirement of our model for GPU memory can be satisfied for megapixelface swapping. In summary, complete face representation, stable training, andlimited memory usage are the three novel contributions to the success of ourmethod. Extensive experiments demonstrate the superiority of MegaFS and thefirst megapixel level face swapping database is released for research onDeepFake detection and face image editing in the public domain. The dataset isat this link.