macrodata/wasup_homer_gold

WASUP HomER黄金标注是一个小型的人工编辑时间性子任务标注集，用于公共HomER自我中心视频。数据集以Parquet文件格式提供，包括轻量级版本（其中video字段为上游HomER视频URL）和嵌入版本（其中video字段为直接存储在Parquet文件中的原始MP4字节）。每个行代表一个视频片段，包含id（标注工具使用的片段标识符）、video（视频URL或MP4字节）、instruction（归一化的片段级别指令）和segments（时间性子任务标注列表，每个标注包括start_sec开始时间、end_sec结束时间和annotation标注内容）。文本规范化处理涉及去除首尾空格并将首字母小写。WASUP HomER Gold Annotations is a small human-edited temporal subtask annotation set for public HomER egocentric videos. The dataset is provided in Parquet files, including a lightweight version where the video field is the upstream HomER video URL and an embedded version where the video field is raw MP4 bytes stored directly in the Parquet file. Each row represents one video episode with id (episode id used by the annotation tool), video (HomER URL or MP4 bytes), instruction (normalized episode-level instruction), and segments (a list of temporal subtask annotations with start_sec, end_sec, and annotation). Text normalization involves stripping leading/trailing whitespace and lowercasing the first character.