Shijia2025/Struggle-Dataset

Struggle Determination Datasets是一个用于视频分类任务的数据集，专注于从第一人称视角检测手-物体交互过程中的挣扎动作。该数据集包含三个子数据集：Pipes-Struggle、Tent-Struggle和Tower-Struggle，分别对应组装水管、搭建帐篷和玩汉诺塔游戏的任务。数据集通过头戴式GoPro相机记录参与者在执行这些任务时的第一人称视频，视频分辨率为1920×1080，帧率为30或60Hz。原始视频被统一修剪为10秒的片段，并调整分辨率为456×256。挣扎动作的标注由MTurk工作者和人类专家共同完成，标注内容包括挣扎程度的不同等级。数据集的结构包括注释文件、视频片段、提取的帧图像以及用于数据处理和分割的工具脚本。

The Struggle Determination Datasets are designed for video classification tasks, focusing on detecting struggle actions during hand-object interactions from an egocentric view. The dataset comprises three subsets: Pipes-Struggle, Tent-Struggle, and Tower-Struggle, corresponding to tasks of assembling plumbing pipes, pitching a tent, and playing the Tower of Hanoi game, respectively. First-person videos were recorded using a head-mounted GoPro camera as participants performed these tasks, with a resolution of 1920×1080 and frame rates of either 30 or 60Hz. The raw videos were uniformly trimmed into 10-second segments and resized to 456×256 resolution. Struggle determination annotations were provided by both MTurk workers and human experts, rating the videos on different levels of struggle. The dataset structure includes annotation files, video segments, extracted frames, and tools for data processing and splitting.