Meta Platforms, Inc.
Color sampling selection for displaying content items using machine learning

Abstract:

An online system is configured to provide content items to users. The content item includes an image, and is displayed with an interface element colored using an accent color to create a unified look and feel with the displayed image. The accent color is dynamically selected based upon the image, extracted color features of the image, and embeddings associated with the image indicating at least one object depicted in the image. A machine-trained classification model selects the color to be used in displaying the interface element from a quantized set of colors of the image, based upon the extracted color features and the embeddings associated with the image. As such, suitable accent colors can be selected automatically for large numbers of content items, in a flexible manner that can account for the context of the images and the context in which the content item is to be displayed.