CMU Panoptic (CMU Panoptic Dataset)

CMU Panoptic是一个大规模数据集，为多个人参与社交活动提供3D姿势注释 (1.5百万)。它包含65个视频 (5.5小时) 与多视图注释，但其中只有17个是在多人场景和相机参数。
大规模多视图系统
480 VGA相机视图
30个高清视图
10个RGB-D传感器
基于硬件的同步
校准
带有标签的有趣场景
多人
社会互动群体
3D身体姿势
3D面部地标
成绩单演讲者ID
硬件设置
480 VGA相机，640x480分辨率，25 fps，使用硬件时钟相互同步
31个高清摄像机，1920x1080分辨率，30 fps，使用硬件时钟相互同步，与VGA摄像机对齐的时间
10个Kinect Ⅱ 传感器。1920x1080 (RGB)，512x424 (深度)，30 fps，它们与其他传感器之间的定时对齐5 DLP投影仪。与高清摄像机同步

CMU Panoptic is a large-scale dataset that provides 3D pose annotations (totaling 1.5 million) for social activities involving multiple people. It contains 65 videos (5.5 hours in total) with multi-view annotations, but only 17 of them cover multi-person scenarios and complete camera parameter configurations.

Large-scale multi-view system:
480 VGA camera views
30 high-definition (HD) camera views
10 RGB-D sensors
Hardware-based synchronization
Calibration
Annotated engaging scenarios
Multiple people
Social interaction groups
3D body poses
3D facial landmarks
Speaker ID transcriptions

Hardware setup:
480 VGA cameras with 640×480 resolution, 25 fps, synchronized with one another via hardware clocks
31 HD cameras with 1920×1080 resolution, 30 fps, synchronized via hardware clocks and temporally aligned with the VGA cameras
10 Kinect II sensors: 1920×1080 (RGB) and 512×424 (depth) resolutions, 30 fps, with temporal alignment to other sensors; 5 DLP projectors synchronized with the HD cameras

搜集汇总

背景与挑战

背景概述

CMU Panoptic是一个大规模多视图数据集，专注于多人社交活动的3D姿势注释，包含65个视频（5.5小时）和480个VGA相机等多传感器数据。由卡内基梅隆大学和北京邮电大学于2017年发布，适用于社交互动和3D身体姿势研究。