STAR

The 'STAR' dataset/benchmark is used to evaluate multimodal models' performance in video question answering by providing a structured set of video data with associated questions and answers.

Papers using STAR (3)

BoxTuning: Directly Injecting the Object Box for Multimodal Model Fine-Tuning2026Clue Matters: Leveraging Latent Visual Clues to Empower Video Reasoning2026Cotasks: Chain-of-thought Based Video Instruction Tuning Tasks2025