Safety Helmet detection with Extended Labels (SHEL)

We extended the number of labels in Kaggle’s safety helmet detection dataset, which has 5000 images and 5000 annotations. The original dataset had three classes (person, head and helmet) and a total of 2501 labels. Moreover, the original dataset was incompletely labelled. We added three new labels on the dataset in results, the new labels consists of six classes (helmet, head with helmet, person with helmet, head, person no helmet, and face) and total of 75578 labels.

我们对Kaggle平台的安全帽检测数据集的标注规模进行了扩充。该原始数据集包含5000张图像与5000个标注框，原本仅设有3个类别：人员（person）、头部（head）与安全帽（helmet），总计2501个标注标签，且存在标注不完整的问题。我们为该数据集新增了3个类别，优化后的标注体系共包含6个类别：安全帽（helmet）、佩戴安全帽的头部（head with helmet）、佩戴安全帽的人员（person with helmet）、头部（head）、未佩戴安全帽的人员（person no helmet）以及面部（face），总标注标签数量达到75578个。