CrossCult-KIBench-sample
收藏资源简介:
CrossCult-KIBench Sample 是一个用于评审检查、轻量级下载和快速格式检查的确定性5%样本数据集,不应用于最终基准测试结果的报告。该样本包含880个案例,覆盖英语、中文和阿拉伯语,涉及视觉问答和图像文本到文本的任务。数据集结构包括训练集和测试集(单次插入和顺序插入案例),每个案例包含案例ID、主题、场景名称、图像路径、语言、问题、目标答案等字段。图像部分包含捆绑图像和可重建的第三方衍生图像,共计2,319个唯一图像路径。数据集遵循CC BY-NC 4.0许可,适用于非商业研究用途。样本主要用于格式检查和数据加载测试,不推荐用于文化排名或政策决策等用途。
CrossCult-KIBench Sample is a deterministic 5% sample dataset for review checks, lightweight downloads, and quick format checks, and should not be used for reporting final benchmark results. The sample contains 880 cases covering English, Chinese, and Arabic, involving visual question answering and image-text-to-text tasks. The dataset structure includes training and test sets (single-insert and sequential-insert cases), with each case containing fields such as case ID, topic, scene name, image path, language, question, target answer, etc. The image portion includes bundled images and reconstructable third-party derived images, totaling 2,319 unique image paths. The dataset is licensed under CC BY-NC 4.0 and is suitable for non-commercial research purposes. The sample is primarily intended for format checks and data loading tests and is not recommended for uses such as cultural ranking or policy decision-making.




