Oxford Road Boundaries

Oxford Road Boundaries 是一个数据集，旨在训练和测试基于机器学习的道路边界检测和推理方法。作者已经对来自 Oxford Robotcar 数据集的 10 公里长的尝试中的两个进行了手工注释，并从其他尝试中生成了数千个带有半注释道路边界掩码的进一步示例。为了以这种方式增加训练样本的数量，作者使用基于视觉的定位器将标注数据集的标签投影到不同时间和天气条件下的其他遍历。因此，该数据集由 62,605 个标记样本组成，其中 47,639 个样本是经过管理的。这些样本中的每一个都包含左右镜头的原始掩码和分类掩码。我们的数据包含来自各种场景的图像，例如笔直的道路、停放的汽车、路口等。

Oxford Road Boundaries is a dataset intended for training and testing machine learning-based road boundary detection and reasoning methods. The authors manually annotated two of the 10-kilometer traversals from the Oxford Robotcar dataset, and generated thousands of additional examples with semi-annotated road boundary masks from other traversals. To increase the size of the training sample set in this manner, the authors utilized a vision-based localizer to project the labels from the annotated dataset onto other traversals conducted under varying time and weather conditions. Accordingly, this dataset comprises 62,605 labeled samples, among which 47,639 samples are curated. Each of these samples contains the original masks and classification masks for both the left and right camera lenses. Our dataset includes images captured in diverse scenarios, such as straight roads, parked cars, intersections, and more.

搜集汇总

背景与挑战

背景概述

Oxford Road Boundaries数据集专为道路边界检测和推理方法而设计，基于Oxford Robotcar数据构建，包含62,605个标注样本，其中47,639个经过管理。每个样本提供左右镜头的原始掩码和分类掩码，涵盖多种道路场景，如直线道路、停车车辆和交叉路口。