Algorithm 3: Facial image de-identification.

Input: A human face image $I\in {\Re }^{n\times m\times 3}$ to protect; a pretrained generator $G(·)$   Output: The de-identify facial image $I_d$.   Initialize latent code ${\omega }^{*}=\omega$;   while not converged do $| I\simeq I^{\prime }=G\left({\omega }^{*}\right)$;   end$\phantom{(}$ $I_d=G({\omega }^{*}+Lap(\frac{\mathsf{\Delta }f}{ϵ}))$