自动驾驶多模态多场景大规模数据集

该数据集是一个公共大规模数据集，在波士顿和新加坡收集了1000个驾驶场景，这两个城市以其浓厚的交通和极具挑战性的驾驶情况而闻名。手动选择20秒长度的场景，为了促进常见的计算机视觉任务，例如对象检测和跟踪，注释了23个对象类，并在整个数据集上的2Hz处具有准确的3D边界框。另外，我们注释对象级属性，例如可见性，活动和姿势。完整的数据集包括大约140万个相机图像，390k激光雷达，140万雷达扫描和1.4m的40k密钥框架对象边界框。

This is a public large-scale dataset containing 1,000 driving scenarios collected in Boston and Singapore—two cities renowned for their heavy traffic and highly challenging driving conditions. All selected 20-second-long scenarios are manually curated. To support common computer vision tasks such as object detection and tracking, 23 object classes are annotated with accurate 3D bounding boxes at 2Hz across the entire dataset. Additionally, object-level attributes including visibility, activity status and pose are annotated. The full dataset comprises approximately 1.4 million camera images, 390,000 LiDAR scans, 1.4 million radar scans, and 40,000 key frames that collectively hold 1.4 million object bounding boxes.

搜集汇总

背景与挑战

背景概述

该数据集是一个公共大规模自动驾驶多模态数据集，在波士顿和新加坡收集了1000个驾驶场景，涵盖约140万相机图像、39万激光雷达和140万雷达扫描。它提供了23个对象类的3D边界框注释，并包括对象级属性标注，以支持计算机视觉任务如对象检测和跟踪。