ONCE

该数据集名为ONCE，是一个大规模的自动驾驶数据集，包含了100万个激光雷达点云样本，这些样本对应于在中国不同城市累计的144小时驾驶数据。该数据集旨在评估半监督学习和自监督学习方法在三维目标检测领域的性能。数据集包含了581个序列，其中20个为标注序列，561个为未标注序列。标注数据被划分为训练、验证和测试集，覆盖了多种天气条件。性能评估采用基于三维交并比（IoU）阈值的平均精度（mAP）指标。该数据集的规模达到了100万样本，任务重点在于三维目标检测。

This dataset, named ONCE, is a large-scale autonomous driving dataset containing 1 million LiDAR point cloud samples, which correspond to 144 hours of driving data accumulated across different cities in China. It is designed to evaluate the performance of semi-supervised learning and self-supervised learning methods in the domain of 3D object detection. The dataset includes 581 sequences, among which 20 are labeled sequences and 561 are unlabeled ones. The labeled data is split into training, validation, and test sets, covering various weather conditions. Performance assessment adopts the mean Average Precision (mAP) metric based on 3D Intersection over Union (IoU) thresholds. With a scale of 1 million samples, the primary task of this dataset is 3D object detection.