A novel framework that extracts object-behavior facts from video clips, reasons over those facts using transformers, and predicts the adverb types that best describe the overall video content.
