Semantically-Prompted Language Models Improve Visual Concept Descriptions
Leveraging semantic knowledge bases and contrastive prompting, V-GLOSS generates detailed and distinguishing visual descriptions that improve performance on zero-shot vision tasks.
