HICO-DET

该数据集名为HICO-DET，包含了38,118张用于训练的图片和9,658张用于测试的图片。它涵盖了80个物体类别、117个交互类别以及600个人与物体交互（HOI）类别。在默认和已知物体设置下，该数据集通过全量、罕见和非罕见划分来评估，并报告了mAP值。其规模包括38,118张训练图片和9,658张测试图片，任务重点在于人与物体交互的检测。

The dataset is named HICO-DET, which contains 38,118 training images and 9,658 test images. It covers 80 object categories, 117 interaction categories, and 600 human-object interaction (HOI) categories. Under the default and known-object settings, this dataset is evaluated via full, rare, and non-rare splits, and the mAP values are reported. With a scale of 38,118 training images and 9,658 test images, its task focuses on human-object interaction detection.

搜集汇总

背景与挑战

背景概述

HICO-DET是一个专注于人与物体交互（HOI）识别的基准数据集，支持分类和检测两种任务，包含多样化的交互类别和详细的实例标注。该数据集与HICO共享HOI类别，并且在图像数量上更为全面。