Face Swapping with Limited Input

We propose an end-to-end face swapping pipeline. To replace the faces in a video, we take in a single photo to produce the result, while previous methods either need sophisticated equipment or large amount of data. Our pipeline consists of three parts, expression transfer, details refining and merging back. By combing 3D model and Generative Adversarial Network, we show that our pipeline is easy to implement, and is able to produce decent results even with limited input.