Meta Platforms, Inc.
Detecting an identity of a brand and/or an obfuscated identity of the brand within an image based on embeddings corresponding to regions of interest within the image

Abstract:

An online system receives an image associated with a set of pixel values and provides the set of pixel values to a deep learning or machine-learned model that is trained to assign a set of regions of interest within the image to one or more classes. The model is structured to identify the set of regions of interest based on the pixel values, in which each region of interest is associated with an embedding corresponding to a vector representation of the region of interest. The online system compares each embedding to a database of embeddings associated with multiple labeled images, in which the labels indicate an identity of a brand and/or an obfuscated identity of the brand. The online system then determines whether the identity of the brand and/or the obfuscated identity of the brand is/are detected within each region of interest based on the comparison.