LCric

LCric是由印度理工学院罗克分校、普林斯顿大学和多伦多大学的研究团队开发的一个大规模长视频理解基准数据集。该数据集包含超过1000小时的密集注释板球视频，平均样本长度约为50分钟。通过自动化注释和视频流对齐管道（ASAP），研究团队能够以几乎零注释成本收集这些视频。LCric数据集不仅用于评估视频理解模型，还通过大量组合多选择和回归查询来分析这些模型的性能。此外，该数据集还用于建立人类基准，表明在长视频理解方面仍有显著的研究空间。

LCric is a large-scale long-form video understanding benchmark dataset developed by research teams from the Indian Institute of Technology Roorkee, Princeton University, and the University of Toronto. The dataset includes over 1,000 hours of densely annotated cricket videos, with an average sample duration of approximately 50 minutes. Leveraging the Automated Annotation and Stream Alignment Pipeline (ASAP), the research team collected these videos at nearly zero annotation cost. The LCric dataset is not only used to evaluate video understanding models, but also to analyze their performance via a large set of combined multiple-choice and regression-based queries. Additionally, this dataset is employed to establish human performance baselines, demonstrating that there remains significant research potential in the domain of long-form video understanding.