Youtube-Infant-Body-Parsing Video Dataset

该数据集包括由Chambers等人 (2020) 从Youtube收集的90个婴儿运动视频。视频的url在其github中提供。由于原始视频的长度很长，我们下载了视频并每4帧对其进行下采样。采样帧可在此google驱动器上使用。每个帧名的格式为 {7_digits_of_video_index}{6_digits_of_frame_index}。对于婴儿身体解析，我们收集了五个类别的注释: 背景，头部，手臂，躯干和腿部。我们将所有视频随机拆分为68个训练视频和22个测试视频，产生2,149个标记和4,690个未标记的训练帧，1,256个标记和2,737个未标记的测试帧。

This dataset contains 90 infant movement videos collected from YouTube by Chambers et al. (2020). The URLs of these videos are available on their GitHub repository. Given the considerable length of the original videos, we downloaded them and performed downsampling by extracting every 4th frame. The downsampled frames are accessible on this Google Drive. The naming format of each frame is {7-digit video index}{6-digit frame index}. For infant body parsing, we collected annotations for five categories: background, head, arms, torso, and legs. We randomly split all videos into 68 training videos and 22 test videos, resulting in 2,149 labeled training frames and 4,690 unlabeled training frames, as well as 1,256 labeled test frames and 2,737 unlabeled test frames.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于婴儿身体解析的视频数据集，包含90个从Youtube收集的婴儿运动视频，经过每4帧下采样处理，并提供了五个类别（背景、头部、手臂、躯干、腿部）的注释。数据集由约翰霍普金斯大学、宾夕法尼亚州立大学和暨南大学于2022年发布，已划分为训练集和测试集，适用于计算机视觉和婴儿行为分析研究。