大规模自动驾驶数据集

该数据集包含在波士顿和新加坡采集的约1000个20秒场景，涵盖28130个训练、6019个验证和6008个测试样本，注释频率为2Hz。配备32光束激光雷达、6摄像头及360°雷达，支持3D对象检测，评估汽车、行人、自行车等10类目标。适用于自动驾驶感知系统训练与评测。

This dataset comprises approximately 1,000 20-second scenarios collected in Boston and Singapore, with 28,130 training samples, 6,019 validation samples, and 6,008 test samples, and an annotation frequency of 2 Hz. It is equipped with a 32-beam LiDAR, six cameras, and a 360° radar, supporting 3D object detection tasks for 10 target categories including cars, pedestrians, bicycles and other common road objects. This dataset is intended for the training and evaluation of autonomous driving perception systems.

搜集汇总

背景与挑战

背景概述

nuScenes是一个大规模自动驾驶数据集，采集于波士顿和新加坡，包含约1000个20秒场景，总计超过4万个样本，注释频率为2Hz。数据集配备多模态传感器（32光束激光雷达、6摄像头和360°雷达），支持3D对象检测，评估汽车、行人、自行车等10类目标，适用于自动驾驶研究和开发。