luuuulinnnn/Nurec_eval_v2
收藏资源简介:
Nurec_eval_v2数据集包含59个OOD(分布外)长尾场景,采用MADS模式,与trajdata兼容。该数据集专为自动驾驶闭环评估设计,提供高清地图数据(如车道、道路边界、等待线、交通标志等),适用于NVIDIA的AlpaSim仿真平台。数据集中的场景来自8个OOD长尾类别,包括动物/鸟类/道路伤亡、复杂交叉口交互、骑行者和微移动设备复杂场景、紧急事件场景、行人密集/近距离场景、道路碎片/安全痕迹、特殊/不常见车辆行为以及工作区/临时交通控制。每个场景以USDZ格式打包,包含场景元数据、高斯重建权重、网格数据、记录轨迹和地图数据文件。数据集是Nurec_eval v1的升级版,改进了地图模式兼容性,确保AlpaSim能自动加载地图并作为场景上下文输入给驾驶模型。
Nurec_eval_v2 is a dataset containing 59 OOD (Out-of-Distribution) longtail scenes in the MADS schema, compatible with trajdata. It is designed for closed-loop evaluation in autonomous driving, providing HD map data (e.g., lanes, road boundaries, wait lines, traffic signs) for use with NVIDIAs AlpaSim simulation platform. The scenes are categorized into 8 OOD-longtail buckets, including Animals/Birds/Roadkill, Complex Intersection Interaction, Cyclists & Micromobility Complex, Emergency Incident Scene, Pedestrian Density/Close Proximity, Road Debris/Safety Traces, Special/Uncommon Vehicle Behavior, and Work Zones/Temp Traffic Control. Each scene is packaged in USDZ format, containing scene metadata, Gaussian-splat reconstruction weights, mesh data, recorded trajectories, and map data files. This dataset is an upgraded version of Nurec_eval v1, with improved map schema compatibility, ensuring AlpaSim can auto-load the map and feed it as scenario context to driver models.





