| Popis: |
First impressions are influential in shaping our personal, economic, and political decisions. We develop a computational framework that can model and modify impressions of faces. First, we use a state-of-the-art predictive model of facial impressions (such as facial attractiveness, trustworthiness, and intelligence) and apply it to a large-scale natural face dataset in order to create a robust facial impression dataset. We validate the augmented dataset with respect to human judgments. Second, we use the new dataset to train a model, ModifAE, that changes face smoothly and effectively in multiple social dimensions. This modification model offers social scientists the ability to manipulate impressions as needed, and it sheds light on both the biases and the visual features underlying first impression formation. |
