MPHOI-72

MPHOI-72是可用于各种HOI/活动识别和姿势估计/对象跟踪任务的多人对象交互数据集。由于人类和物体之间的许多身体遮挡，该数据集具有挑战性。它由从30 fps的3个不同角度捕获的72个视频组成，总共26,383帧，平均长度为12秒。它涉及5个人成对表演，6个对象类型，3个活动和13个子活动。数据集包括彩色视频，深度视频，人体骨骼，人和物体边界框。MPHOI-72 is a multi-person human-object interaction (HOI) dataset suitable for a wide range of HOI/activity recognition, pose estimation and object tracking tasks. This dataset is challenging due to extensive physical occlusions between humans and objects. It consists of 72 videos captured from 3 distinct perspectives at 30 fps, with a total of 26,383 frames and an average duration of 12 seconds. It involves five paired human performers, 6 object categories, 3 primary activities and 13 sub-activities. The dataset provides color videos, depth videos, human skeletons, as well as bounding boxes for both humans and objects.

搜集汇总

背景与挑战

背景概述

MPHOI-72是一个多人对象交互数据集，包含从3个不同角度捕获的72个视频，共26,383帧，涉及5个人、6个对象类型和多种活动。数据集提供彩色视频、深度视频、人体骨骼和边界框等丰富数据，适用于HOI、活动识别和姿势估计等任务。