Ceteris Paribus Face Challenge : Team ArgoCS

Description of our methods for the Ceteris Paribus Face Challenge.
The Ceteris Paribus Face Challenge was an AI competition organized by Jules Salzinger and Nicolas Posocco, consisting in the modification of a given FFHQ subset. The modification consisted, for a set of characteristics, to make a "copy" of each image for each of their missing characterictics.
e.g. If face X is blond, make a brown hair, black hair, white hair and bald.
The basis of our method consisted in inverting GAN models to influence the images' latent space representations using existing implementation Encoder4Editing. Using efficient GUIs and some tricks (described in the video :), we managed to reach 2nd place on the main leaderboard.
Our method was ranked 1st among the 3 participating teams, and our results 2nd, averaging to a 2nd place on mainleaderboard.
Presentation by all the members of the team:
Hédi Razgallah
Thibault LSDC