MMAct

该数据集名为MMAct，是一个大规模的数据集，包含了约1900个序列，涵盖了35个动作类别，来自40位参与者在7种不同模态下的数据。由于数据收集过程中场景和用户朝向的多样性，该数据集在动作识别任务中提出了较大的挑战。其规模属于大型，所针对的任务是人类动作识别。The dataset named MMAct is a large-scale dataset that includes approximately 1,900 sequences covering 35 action categories, with data collected from 40 participants across 7 distinct modalities. Given the diversity of scenarios and user orientations during data collection, this dataset poses substantial challenges for human action recognition tasks. As a large-scale dataset, it is specifically designed for the task of human action recognition.

搜集汇总

背景与挑战

背景概述

MMAct是一个用于人类动作理解的大规模跨模态学习数据集，包含7种模态（如RGB、关键点、传感器数据等）、1900多个视频和40k+个动作实例，覆盖37个日常和异常动作类别，并在5个不同场景中由40名受试者通过半自然协议收集，支持多视角分析。