CODA desc 和 nuScenes desc
收藏资源简介:
CODA desc 和 nuScenes desc 是由同济大学上海智能自主系统研究所创建的两个用于交通场景语义理解的新数据集。这些数据集旨在为每个驾驶场景提供丰富且无幻觉的语义描述,支持未来的研究工作。数据集通过HCOENet方法创建,该方法结合了幻觉交叉检查框架和关键对象增强框架,确保描述的准确性和完整性。数据集的应用领域主要集中在自动驾驶系统的语义场景理解,旨在解决自动驾驶系统中的幻觉问题,提高驾驶安全性。
CODA desc and nuScenes desc are two novel datasets for traffic scene semantic understanding created by the Shanghai Institute of Intelligent Autonomous Systems, Tongji University. These datasets aim to provide rich and hallucination-free semantic descriptions for each driving scenario, supporting future research work. The datasets are developed via the HCOENet method, which combines a hallucination cross-checking framework and a key object enhancement framework to ensure the accuracy and completeness of the descriptions. Their application fields mainly focus on semantic scene understanding for autonomous driving systems, aiming to solve the hallucination problem in autonomous driving systems and improve driving safety.

- 1Hallucination Elimination and Semantic Enhancement Framework for Vision-Language Models in Traffic Scenarios上海智能自主系统研究所,同济大学 · 2024年



