StreetHazards 异常检测数据集

StreetHazards 是面向异常物检测的合成图像数据集，由研究人员使用虚拟引擎和 CARLA 模拟环境创建而成。研究人员在驱动场景中插入各种各样外来物体，并用这些新物体重新渲染创造场景。数据集中有 5,125 个图像和语义分割 ground truth 组用于训练，1,031 个无异常值组用于验证，1,500 个有异常值组用于测试。训练集共有 12 类，分别是背景、道路、街道线、交通标志、人行道、行人、车辆、建筑、墙、杆子、栅栏、植被。

StreetHazards is a synthetic image dataset designed for anomaly detection, created by researchers using game engines and the CARLA simulation environment. Researchers insert various foreign objects into driving scenarios and re-render the scenes to generate samples with these new objects. The dataset includes 5,125 pairs of images and semantic segmentation ground truth for training, 1,031 outlier-free pairs for validation, and 1,500 outlier-containing pairs for testing. The training set covers 12 categories, namely background, road, street lines, traffic signs, sidewalks, pedestrians, vehicles, buildings, walls, poles, fences, and vegetation.

搜集汇总

背景与挑战

背景概述

StreetHazards是一个用于异常物检测的合成图像数据集，包含5,125个训练图像和1,500个测试图像，共12个类别。该数据集通过虚拟引擎和CARLA模拟环境创建，主要用于目标检测任务。