Detecting objects in images using arbitrary text descriptions rather than a fixed set of predefined categories.
Quality of vision, audio, and image understanding (distinct from modality support)