REXTIME
收藏资源简介:
REXTIME数据集由国立台湾大学开发,是一个专注于视频时间推理的基准,特别强调在不同视频片段间进行因果关系理解的能力。该数据集包含2143个精心策划的测试样本,每个样本都经过人工标注以确保准确性和相关性。数据集的创建过程利用了大型语言模型辅助的数据生成管道,显著减少了人工标注的需求。REXTIME主要应用于评估和提升多模态AI模型在视频时间推理方面的能力,特别是在机器人、医疗和法律政策制定等领域。
The REXTIME dataset, developed by National Taiwan University, is a benchmark focused on video temporal reasoning, with a particular emphasis on the ability to understand causal relationships across different video segments. This dataset contains 2143 carefully curated test samples, each of which has been manually annotated to ensure accuracy and relevance. The dataset's creation process leverages a Large Language Model-aided data generation pipeline, which significantly reduces the need for manual annotation. REXTIME is primarily used to evaluate and enhance the capabilities of multimodal AI models in video temporal reasoning, especially in fields such as robotics, healthcare, and legal policy-making.




