论文总结：Deepfake

(non-detailed summary about Deepfake)

• normally deepfake is to generate/control humans images. for example, take on clothes during virtual shopping, change actors’ face and so on.

• deepfake consists of reenactment, replacement, editing and synthesis.

  • reenactment is to driver xt(target) by xs(source) to do the same action as xs. it includes expression reenactment, mouth reenactment(lip synthesis), eyes reenactment(gaze), head reenactment(head-on), body reenactment.
  • replacement is to replace xt with xs. It includes transfer(face transfer), swap(change the normal face with famous face).
  • editing is to add, change or delete the characters of xt.
  • synthesis is to create a character without xt. It can used to create movie’s or animation’s characters

• deepfake can be implemented through in followed methods

  • encoder-decoder networks (ED)