MPII Human Pose
收藏资源简介:
MPII 人体姿势数据集是用于评估关节式人体姿势估计的最先进的基准。该数据集包括大约 25K 幅图像,其中包含超过 40K 人带有注释的身体关节。这些图像是使用已建立的日常人类活动分类法系统地收集的。总体而言,该数据集涵盖了 410 个人类活动,并且每个图像都带有一个活动标签。每个图像都是从 YouTube 视频中提取的,并提供了前后未注释的帧。此外,对于测试集,我们获得了更丰富的注释,包括身体部位遮挡和 3D 躯干和头部方向。遵循文献中性能评估基准的最佳实践,我们保留测试注释以防止过度拟合和调整测试集。我们正在开发基于丰富测试集注释的自动评估服务器和性能分析工具。
The MPII Human Pose Dataset is a state-of-the-art benchmark for evaluating articulated human pose estimation. This dataset contains approximately 25K images, with over 40K annotated human body joints for more than 40K individuals. These images were systematically collected using a well-established taxonomy of daily human activities. Overall, the dataset covers 410 human activities, and each image is associated with an activity label. Each image is extracted from a YouTube video, with unannotated frames before and after the extracted image provided. Furthermore, for the test split, we obtained richer annotations, including body part occlusion information and 3D trunk and head orientations. Following the best practices for performance evaluation benchmarks in academic literature, we withheld the test annotations to prevent overfitting caused by tuning models on the test set. We are currently developing an automatic evaluation server and performance analysis tools based on the rich annotations from the test split.

- MPII Human Pose数据集首次发表,由Andreas Andriluka等人在德国马克斯·普朗克信息学研究所创建,旨在为人体姿态估计研究提供一个全面且高质量的数据集。
- MPII Human Pose数据集首次应用于学术研究,成为人体姿态估计领域的重要基准,推动了相关算法的发展和评估。
- 随着深度学习技术的进步,MPII Human Pose数据集被广泛用于训练和测试各种基于卷积神经网络的人体姿态估计模型,显著提升了模型的性能。
- MPII Human Pose数据集的扩展版本发布,增加了更多的标注数据和多样化的场景,进一步丰富了数据集的内容和应用范围。
- MPII Human Pose数据集在多个国际计算机视觉竞赛中被用作标准测试集,验证了其在人体姿态估计领域的持续影响力和重要性。



