HOD-Benchmark-Dataset

本研究提出了一个名为HOD-Benchmark-Dataset的新基准数据集，用于有害物体检测，由延世大学等机构创建。该数据集包含超过10,000张图像，涵盖6个可能有害的类别，不仅包括正常情况，还包括难以检测的困难情况。数据集的创建过程涉及使用搜索关键词收集图像，并通过多标签分类方式进行标注。该数据集主要应用于在线服务中自动检测和过滤有害视觉内容，旨在减少对用户的潜在负面影响。

This study proposes a novel benchmark dataset named HOD-Benchmark-Dataset for harmful object detection, which was developed by Yonsei University and other institutions. This dataset contains over 10,000 images covering 6 categories of potentially harmful content, including both normal scenarios and hard-to-detect edge cases. The dataset curation process involves collecting images via search keywords and annotating them through multi-label classification. This dataset is primarily applied to automatically detect and filter harmful visual content in online services, aiming to mitigate potential negative impacts on users.