2/20/2023 0 Comments Dying light at target![]() This submission surfaces internal biases of the core component of the target image-cropping algorithm (saliency prediction). Second, we use qualitative thematic analysis of the obtained tweaked images to show that the target model is biased towards deeming more "salient" the depictions of people that appear slim, young, of light or warm skin color and smooth skin texture, and with stereotypically feminine facial traits. This generates a dataset of images tweaked to make the target saliency-prediction model more excited. First, using a large-scale computer-vision model, we manipulate an image of a person in a way that increases maximum saliency predicted by the target model. ![]() In this submission we present and apply a two-step mixed method for analyzing the bias of internal representations of the target saliency-prediction model. ![]() This is a submission to the Twitter Algorithmic bias How to Become More Salient? Surfacing Representation Biases of the Saliency Prediction Model ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |