StreetHazards

该数据集是一个大规模的数据集，包含了不同系列的街道场景合成图像，旨在评估认识不确定性以及检测分布外数据。在训练数据集中，包含了13个类别的标注信息，而测试数据集中不仅包含这些类别，还额外包含了250个未见过的分布外类别。数据集的规模为：训练图像5,125张，测试图像1,500张。其任务旨在进行分布外检测以及语义分割。

This is a large-scale dataset composed of synthetic street scene images across different series, designed for evaluating epistemic uncertainty and out-of-distribution (OOD) detection tasks. The training dataset provides annotated labels for 13 categories, while the test dataset covers these 13 categories plus an additional 250 unseen out-of-distribution categories. The dataset has 5,125 training images and 1,500 test images in total. Its core tasks include out-of-distribution detection and semantic segmentation.