MPII Human Pose Estimation Dataset

MPII is a dataset and related benchmarks for evaluating human pose estimation, with approximately 25,000 images and containing more than 40,000 people with annotated joints. The dataset systematically collects images using an established taxonomy of human activities.

Overall, the dataset covers 410 human actions and provides activity labels for each image. Each image comes from a YouTube video and provides a related unannotated frame. In addition, the annotations of the test set include body part occlusion, 3D torso, and head orientation.

This dataset was released by the Max Planck Institute for Informatics in 2014, and the related paper is "2D Human Pose Estimation: New Benchmark and State of the Art Analysis".