Algorithm 1.

Training ShapeEditor using gradient descent.

Input:      Iattr: Image containing attribute information      Iid: Image containing identity information      P: Identity-attribute image pair space Functions:      Encoder ShapeEditor:P $\to W+$      Generator G$:W+\to I$      Loss ← $L_{\mathrm{\text{id}}}:$ Calculate the identity loss between Iid and Iout.      Loss ← $L_{\mathrm{\text{attr}}}:$ Calculate the attribute loss between Iattr and Iout.      Loss ← $L_{\mathrm{\text{rec}}}:$ Calculate the reconstruction loss between Iid(Iattr) and Iout. Output:      I: Image space      $W+:$ Potential vector space of StyleGAN      Iout: Synthesized face-swapping image

1:       for number of training iterations do:

2:            for Iid, Iattr randomly selected in training dataset do:

3:                 Generate the $W\mathbf{+}$ space vector using [Iid, Iattr]

4:                      ShapeEditor:P $\to W+$

5:                 Generate the face-swapping image Iout using the $W+$ space vector

6:                           G$:W+\to I$

7:                 Calculate the identity loss $L_{\mathrm{\text{id}}}$, the attribute loss $L_{\mathrm{\text{attr}}}$, and the reconstruction loss $L_{\mathrm{\text{rec}}}$

8:                 Update ShapeEditor with loss

9:            end

10:  end.