Researchers used a pre-trained off-the-shelf model from Nvidia.
Digital imagery tagging provides negative context for non white people.
