Highway-Text和Urban-Text
收藏资源简介:
本研究贡献了两个场景描述数据集:Highway-Text和Urban-Text。Highway-Text数据集包括来自下一代模拟(NGSIM)数据集和高速公路无人机数据集(HighD)的场景描述;Urban-Text数据集则包含来自澳门连接自动驾驶(MoCAD)数据集和阿波罗景观(ApolloScape)的场景描述。这些数据集覆盖了各种交通场景,旨在提高LLMs对复杂交通场景的理解,并减少虚假现象。通过使用CoT提示技术,数据集指导LLMs逐步生成上下文感知的语义注释。
This study contributes two scenario description datasets: Highway-Text and Urban-Text. The Highway-Text dataset comprises scenario descriptions sourced from the Next Generation Simulation (NGSIM) dataset and the Highway Drone Dataset (HighD), while the Urban-Text dataset contains scenario descriptions from the Macao Connected Autonomous Driving (MoCAD) dataset and ApolloScape. These datasets cover a wide range of traffic scenarios, with the goal of improving LLMs' understanding of complex traffic scenarios and reducing hallucinations. By utilizing Chain-of-Thought (CoT) prompting techniques, these datasets guide LLMs to generate context-aware semantic annotations step-by-step.




