PEFS: A Validated Dataset for Perceptual Experiments on Face Swap Portrait Videos

Videos obtained by current face swapping techniques can contain artifacts potentially detectable, yet unobtrusive to human observers. However, the perceptual differences between real and altered videos, as well as properties leading humans to classify a video as manipulated, are still unclear. Thus, to support the research on perceived realism and conveyed emotions in face swap videos, this paper introduces a high-resolution dataset providing the community with the necessary sophisticated stimuli. Our recording process has been specifically designed to focus on human perception research and entails three scenarios (text-reading, emotion-triggering, and free-speech). We assess the perceived realness of our dataset through a series of experiments. The results indicate that our stimuli are overall convincing, even for long video sequences. Furthermore, we partially annotate the dataset with noticeable facial distortions and artifacts reported by participants.