harpreetsahota/STONE
收藏资源简介:
STONE是一个大规模多模态数据集,专为越野环境下的3D可通行性预测而设计。该数据集由自主地面车辆(UGV)在韩国的四个户外环境中收集而成,遵循nuScenes格式。它提供了7000个关键帧,包含来自6个摄像头的环绕视图图像(分辨率为1904×1200)、128通道LiDAR扫描(约230,000个点)以及体素级可通行性注释,将地形分类为自由、可通行、潜在可通行和不可通行区域。此外,数据集还包括3D障碍物边界框、自我姿态轨迹以及以约10 Hz频率同步的多传感器数据。此FiftyOne版本是从完整的279个场景集合中分层采样的35个场景(每个场景200帧)组成,组织为分组样本,每个关键帧包含7个切片(6个摄像头和1个LiDAR 3D场景)。数据集支持图像分类和对象检测任务,适用于自动驾驶和机器人导航研究。
STONE is a large-scale multimodal dataset specifically designed for 3D traversability prediction in off-road environments. Collected by unmanned ground vehicles (UGVs) in four outdoor environments across South Korea, it follows the nuScenes data format. The dataset provides 7000 key frames, including surround-view images from 6 cameras with a resolution of 1904×1200, 128-channel LiDAR scans (approximately 230,000 points per scan), and voxel-level traversability annotations that categorize terrain into free, traversable, potentially traversable, and non-traversable regions. Additionally, it includes 3D obstacle bounding boxes, ego-pose trajectories, and multi-sensor data synchronized at a frequency of approximately 10 Hz. This FiftyOne iteration consists of 35 scenes (200 frames per scene) that are hierarchically sampled from the complete 279-scene collection, organized as grouped samples, with each key frame containing 7 slices: 6 camera views and 1 LiDAR 3D scene. The dataset supports image classification and object detection tasks, and is applicable to research in autonomous driving and robotic navigation.




