OCHuman

该数据集名为OCHuman，旨在评估模型对遮挡的鲁棒性，其中包含大量被遮挡的人体实例和复杂的姿态。值得注意的是，该数据集不提供训练集，只包含验证集和测试集。整个数据集共包含4731张图像和8110个人体实例，其任务是进行人体姿态估计。

The dataset named OCHuman is designed to evaluate the robustness of models against occlusion. It contains a large number of occluded human instances and complex human poses. Notably, this dataset does not provide a training split, only including validation and test splits. The entire dataset consists of 4,731 images and 8,110 human instances in total, and the corresponding task is human pose estimation.

搜集汇总

背景与挑战

背景概述

OCHuman数据集是一个专注于高度遮挡人体实例的数据集，包含5081张图像和13360个人体实例，提供边界框、人体姿态和实例掩码的全面注释。其平均MaxIoU为0.573，是目前最复杂和具有挑战性的人体相关数据集之一。