UCLA

We explore multiple ideas on face aging, and we finally settle down on constructing a Face Reconstruction Convolutional Neural Network and a Feature Vector Encoder. Together with a discriminator on age and a discriminator on the distribution of the feature vectors, we are able to generate the face aging transition both forward and backward for a given face, with a known age and a known gender. We make comment on the effectiveness of the discriminators mentioned above, which are included in the model in order to enhance the performance. Our results have shown that the inclusion of both discriminators are effective in different ways.