Dataset for patch-based person classification (person vs. non-person objects) and posture classification (standing vs. sitting vs. squatting). The data was recorded using a Kinect2 sensor and consists of labeled depth image patches of 27 persons in various postures and of various non-person objects. In total, the dataset consists of more than 235,000 samples divided into non-overlapping subsets for training, validation, and test.
https://www.tu-ilmenau.de/neurob/data-sets-code/depth-multi-task/