MSRDailyActivity3D

数据集是使用 Kinect 设备捕获的。有16种活动：喝酒、吃饭、看书、打电话、写纸、使用笔记本电脑、使用吸尘器、振作起来、静坐、扔纸、玩游戏、躺在沙发上、走路、弹吉他、站立起来，坐下。有10个科目。每个受试者执行每个活动两次，一次是站立姿势，一次是坐姿。场景中有一张沙发。记录三个通道：深度图 (.bin)、骨架关节位置 (.txt) 和 RGB 视频 (.avi)。每个通道有 16*10*2=320 个文件。总共有 320*3=960 个文件。请注意，RGB 通道和深度通道是独立记录的，因此它们没有严格同步。
骨架文件的格式如下。第一个整数是帧数。第二个整数是关节数，始终为 20。对于每一帧，第一个整数是行数。当这一帧中恰好检测到一个骨架时，该整数为 40。当没有检测到骨架时为零。检测到两个骨架时为 80（在这种情况下很少见，我们只是在实验中使用第一个骨架）。对于大多数帧，行数为 40。每个关节对应两行。第一行是它的真实世界坐标 (x,y,z)，第二行是它的屏幕坐标加上深度 (u, v, depth)，其中 u 和 v 被归一化到 [0,1] 内。对于每一行，最后的整数应该是置信度值，但它没有用。

This dataset was captured using a Kinect device. It includes 16 distinct activities: drinking, eating, reading books, making phone calls, writing on paper, using a laptop, using a vacuum cleaner, cheering up, sitting still, throwing paper, playing games, lying on the sofa, walking, playing the guitar, standing up, and sitting down. There are 10 total subjects. Each subject performs each of the 16 activities twice: once in a standing posture and once in a sitting posture. A sofa is present in the recording scene. Three modalities are recorded: depth maps (.bin format), skeleton joint positions (.txt format), and RGB video (.avi format). Each modality contains 16*10*2=320 files, resulting in a total of 320*3=960 files across all three modalities. Please note that the RGB and depth channels are recorded independently, so they are not strictly synchronized.

The format of the skeleton data file is specified as follows. The first integer indicates the total number of frames. The second integer specifies the count of skeleton joints, which is consistently 20. For each frame, the first value represents the number of rows in the frame's skeleton data. If exactly one skeleton is detected in the frame, this value is 40; if no skeleton is detected, it is 0; if two skeletons are detected (a rare occurrence in this dataset, only the first skeleton is used for experiments), it is 80. For most frames, the row count is 40. Each joint corresponds to two rows of data. The first row contains the real-world 3D coordinates (x, y, z) of the joint, while the second row contains its screen coordinates paired with depth information (u, v, depth), where u and v are normalized to the range [0, 1]. The final integer in each row is intended to be a confidence score, but it is not actually used.

搜集汇总

背景与挑战

背景概述

MSRDailyActivity3D是一个多模态动作识别数据集，包含16种日常活动的深度图、骨架数据和RGB视频，由10个受试者在不同姿势下执行，总文件数为960个。数据集由多个知名研究机构联合发布，适用于动作识别相关研究。