A learned vector representation of an image that captures its semantic and visual content for retrieval or comparison.