NTU-60

This dataset contains RGB+D videos and skeleton data for human behavior. The behavior data is captured by 3 Microsoft Kinect V2 cameras from 40 human subjects, with a total of 56,880 samples containing 60 categories totaling 4 million frames, where the maximum frame for all samples is 300. 25 joints are recorded for each body skeleton. The dataset provides two original settings, namely two evaluation protocols, Cross-Subject (Xsub) and Cross-View (Xview). In Xsub protocol, the training set contains 40,320 samples from 20 subjects, and the remaining 16,560 samples are used for testing. In Xview protocol, 37,920 samples captured by cameras 2 and 3 are used for training, and camera 1 is used for training. The remaining 18960 samples were used for testing. We follow these two settings and report the Top-1 accuracy of experimental results.本数据集包含用于人类行为研究的RGB+D视频与骨骼数据。该行为数据由3台微软Kinect V2（Microsoft Kinect V2）摄像头采集自40名人类受试者，总计包含56880个样本，涵盖60个行为类别，总帧数达400万，且所有样本的最大帧数为300。每个人体骨骼均记录了25个关节点。本数据集提供两种原生实验设置，即跨受试者（Cross-Subject，Xsub）与跨视角（Cross-View，Xview）两种评估协议。在跨受试者评估协议中，训练集包含来自20名受试者的40320个样本，剩余16560个样本用于测试。在跨视角评估协议中，由摄像头2与3采集的37920个样本用于训练，摄像头1的样本亦用于训练，剩余18960个样本则用于测试。我们遵循这两种实验设置，并报告实验结果的Top-1准确率。

搜集汇总

背景与挑战

背景概述

NTU-60是一个用于人类行为分析的标准数据集，包含RGB+D视频和骨架数据，由3个Kinect V2摄像头从40名受试者捕获，总计56,880个样本覆盖60个行为类别。数据集提供了Cross-Subject和Cross-View两种评估协议，适用于行为识别模型的训练和测试。