Pit30M

Pit30M是由Uber先进技术集团开发的大型数据集，专注于自动驾驶车辆的全局定位问题。该数据集包含超过3000万帧的图像和LiDAR数据，覆盖了一年的驾驶数据，涵盖了多种天气、季节和时间条件。数据集通过自动标注历史天气和天文数据，以及图像和LiDAR的语义分割，提供了精确的定位基准。Pit30M不仅支持图像和LiDAR的检索定位方法研究，还首次为城市尺度下的亚米级检索定位提供了基准。

Pit30M is a large-scale dataset developed by Uber Advanced Technologies Group, focusing on the global localization problem for autonomous vehicles. This dataset contains over 30 million frames of image and LiDAR data, spanning one year of driving recordings and covering diverse weather, seasonal and temporal conditions. It offers precise localization benchmarks via automatically annotated historical weather and astronomical data, coupled with semantic segmentation labels for both images and LiDAR data. Pit30M not only supports research on retrieval-based localization methods using images and LiDAR data, but also provides the first benchmark for sub-meter level retrieval-based localization at the urban scale.