CapRL-Video-178K
收藏资源简介:
CapRL-Video-178K.jsonl 是一个视频路径索引数据集,其核心是一个JSONL格式文件,每条记录包含一个视频文件的相对路径。该数据集本身不包含实际视频文件,而是指向底层视频数据集LLaVA-Video-178K。用户需先下载并解压 `lmms-lab/LLaVA-Video-178K` 中的原始MP4格式视频文件,才能与本数据集的路径配合使用。底层LLaVA-Video-178K数据集总计包含约178K个视频样本,根据视频时长(0-30秒、30-60秒、1-2分钟、2-3分钟)和来源(YouTube、学术来源)被组织成8个子集,各子集样本数量已在README中详细列出。该数据集的主要用途是提供结构化的视频文件路径映射,以方便在视频理解、视频描述生成等需要处理大规模视频数据的机器学习任务中定位和加载对应的视频资源。
CapRL-Video-178K.jsonl is a video path indexing dataset in JSONL format, where each entry contains the relative path of a video file. This dataset does not include the actual video files themselves, but instead points to the underlying video dataset LLaVA-Video-178K. Users must first download and decompress the original MP4-format video files from `lmms-lab/LLaVA-Video-178K` before using the paths in this dataset to locate corresponding video resources. The underlying LLaVA-Video-178K dataset contains approximately 178K video samples in total, organized into 8 subsets based on video duration (0–30 seconds, 30–60 seconds, 1–2 minutes, 2–3 minutes) and source (YouTube, academic sources). The number of samples in each subset is detailed in the accompanying README file. The main purpose of this dataset is to provide structured video file path mappings to facilitate the localization and loading of corresponding video resources in machine learning tasks that require processing large-scale video data, such as video understanding and video caption generation.




