CUlane

CULane是用于交通车道检测学术研究的大规模挑战性数据集。它是由安装在北京不同驾驶员驾驶的六辆不同车辆上的摄像机收集的。收集了超过55小时的视频并提取了133,235帧。数据示例如上所示。我们将数据集分为训练集的88880，验证集的9675和测试集的34680。测试集分为正常和8个具有挑战性的类别，它们对应于上面的9个示例，对于每一帧，我们用三次样条手动注释交通车道。对于车道标记被车辆遮挡或看不见的情况，我们仍根据上下文对车道进行注释，如 (2)(4) 所示。我们还希望算法能够区分道路上的障碍，就像 (1) 中的障碍一样。因此，障碍物另一侧的车道没有注释。在此数据集中，我们将注意力集中在四个车道标记的检测上，这在实际应用中是最受关注的。其他车道标记没有注释。

CULane is a large-scale, challenging dataset for academic research on traffic lane detection. It was collected using cameras mounted on six different vehicles operated by different drivers in Beijing. Over 55 hours of video footage were collected, from which 133,235 frames were extracted. Example data samples are presented above. We partition the dataset into 88,880 training samples, 9,675 validation samples, and 34,680 test samples. The test set is categorized into one normal class and eight challenging classes, which correspond to the nine examples shown above. For every frame, we manually annotate traffic lanes using cubic splines. In cases where lane markings are occluded by vehicles or completely invisible, we still annotate the lanes based on contextual information, as demonstrated in (2) and (4). We also expect that algorithms can distinguish between obstacles on the road, such as the one shown in (1). Accordingly, lanes located on the opposite side of the obstacle are not annotated. In this dataset, we focus on the detection of four lane markings, which are of the utmost concern in real-world applications. No annotations are provided for other lane markings.

AI搜集汇总

背景与挑战

背景概述

CUlane是一个用于交通车道检测的大规模挑战性数据集，包含133,235帧图像，分为训练集、验证集和测试集，特别关注四个车道标记的检测。该数据集由商汤科技和香港中文大学于2018年发布，适用于自动驾驶研究，包含正常和8个具有挑战性的类别，如车辆遮挡和障碍物场景，以评估算法在复杂环境下的性能。