SDT Dataset | Sdt: A Synthetic Multi-Modal Dataset For Person Detection And Pose Classification

The Synthetic Depth & Thermal (SDT) dataset consists of 40k synthetic and 8k real depth and thermal stereo images, depicting human behavior in indoor environments. Included samples show uniquely posed lying, sitting, and standing persons within four different room types (living room, bedroom, bathroom, and kitchen), recorded from an elevated position. Furthermore, a fourth control class with empty rooms is provided as well. Both parts of SDT are balanced sets of these four classes and room types. The synthetic part of the dataset is intended to be used as training (and validation) data for uni-/multi-modal pose classification or person detection models, while the real part can be used to assess the generalization performance. To facilitate supervised training, pose labels and person bounding boxes are provided for all images. The real images in the dataset were captured by a multi-modal stereo camera system, consisting of an Orbbec Astra depth camera and a FLIR Lepton 3.5 thermal camera, while synthetic images, which share the image characteristics of these cameras, were acquired through 3D rendering of virtual scenes within Blender and subsequent introduction of camera-specific noise. Download and Use This data may be used for non-commercial research purposes only. If you publish material based on this data, we request that you include a reference to our paper [1]. [1] C. Pramerdorfer, J. Strohmayer and M. Kampel, "Sdt: A Synthetic Multi-Modal Dataset For Person Detection And Pose Classification," 2020 IEEE International Conference on Image Processing (ICIP), Abu Dhabi, United Arab Emirates, 2020, pp. 1611-1615, doi: 10.1109/ICIP40778.2020.9191284. BibTeX citation: @INPROCEEDINGS{9191284,
author={Pramerdorfer, C. and Strohmayer, J. and Kampel, M.},
booktitle={2020 IEEE International Conference on Image Processing (ICIP)},
title={Sdt: A Synthetic Multi-Modal Dataset For Person Detection And Pose Classification},
year={2020},
volume={},
number={},
pages={1611-1615},
doi={10.1109/ICIP40778.2020.9191284}}

合成深度与热成像（Synthetic Depth & Thermal, SDT）数据集包含4万张合成图像与8千张真实图像，均为深度与热成像立体图像，记录了室内环境中的人类行为场景。样本涵盖四类室内场景（客厅、卧室、浴室与厨房）中，姿态各异的躺卧、端坐与站立人物，所有图像均采用高处视角拍摄。此外，数据集还增设第四类对照样本：空房间场景。数据集的合成与真实两个子集，均在四类人物姿态与四类房间场景上实现样本分布均衡。该数据集的合成子集可作为单模态/多模态姿态分类或人物检测模型的训练（及验证）数据，而真实子集则可用于评估模型的泛化性能。为便于开展监督训练，所有图像均附带姿态标签与人物边界框标注信息。数据集内的真实图像由多模态立体相机系统采集，该系统包含Orbbec Astra深度相机与FLIR Lepton 3.5热成像相机；合成图像则通过在Blender中对虚拟场景进行3D渲染，并后续添加相机专属噪声生成，其图像特性与真实相机采集的图像保持一致。使用与下载须知：本数据集仅可用于非商业性研究用途。若基于本数据集发表研究成果，请务必引用本文献[1]。[1] C. Pramerdorfer、J. Strohmayer与M. Kampel，"Sdt: A Synthetic Multi-Modal Dataset For Person Detection And Pose Classification"，2020 IEEE国际图像处理会议（ICIP），阿联酋阿布扎比，2020年，第1611-1615页，DOI: 10.1109/ICIP40778.2020.9191284。BibTeX引用格式：@INPROCEEDINGS{9191284,
author={Pramerdorfer, C. and Strohmayer, J. and Kampel, M.},
booktitle={2020 IEEE International Conference on Image Processing (ICIP)},
title={Sdt: A Synthetic Multi-Modal Dataset For Person Detection And Pose Classification},
year={2020},
volume={},
number={},
pages={1611-1615},
doi={10.1109/ICIP40778.2020.9191284}}