ROAD-R
收藏资源简介:
ROAD-R数据集是专为自动驾驶设计的,包含22个长约8分钟的视频,每个视频都标注了道路事件。这些道路事件通过一系列时间上相连的边界框(称为tube)来表示,每个边界框都与一组标签相关联,共有41个标签。数据集的目的是预测与每个边界框关联的标签集合。此外,ROAD-R数据集还手动标注了243个逻辑约束,确保模型在预测时遵守这些逻辑规则,从而提高模型的性能和安全性。该数据集的应用领域主要集中在自动驾驶技术中,旨在解决模型在复杂道路环境中如何准确、安全地进行决策的问题。
The ROAD-R dataset is specifically designed for autonomous driving. It includes 22 videos each roughly 8 minutes long, with every video fully annotated with road events. These road events are represented by a series of temporally connected bounding boxes termed tubes, where each bounding box is paired with a set of labels, and there are 41 distinct labels in total. The core objective of this dataset is to predict the label set associated with each individual bounding box. Furthermore, the ROAD-R dataset manually annotates 243 logical constraints to ensure that predictive models adhere to these logical rules, thereby boosting both model performance and safety. The primary application fields of this dataset lie in autonomous driving technology, with the aim of solving the challenge of how models can achieve accurate and safe decision-making in complex road scenarios.




