LLAMAS (Labeled Lane Markers)

通过创建用于自动驾驶的高清地图 (包括基于激光雷达的车道标记) 来注释无监督的Llamas数据集。可以根据这些地图对自动车辆进行定位，并将车道标记投影到相机框架中。通过最小化图像中已经检测到的标记与投影标记之间的差异来优化3D投影。通过使用更好的检测器，优化差异度量并添加一些时间一致性，可能会实现进一步的改进。数据集样本显示在下面的视频中。

This work annotates the unsupervised Llamas dataset by constructing high-definition maps for autonomous driving, which include LiDAR-based lane markings. Autonomous vehicles can be localized using these maps, and lane markings can be projected onto the camera frame. The 3D projection is optimized by minimizing the discrepancy between the markings detected in images and their projected counterparts. Further improvements can be achieved by adopting more advanced detectors, refining the discrepancy metric, and incorporating temporal consistency constraints. Dataset samples are demonstrated in the video below.