NuPlanQA-1M
收藏资源简介:
NuPlanQA-1M是由普渡大学和丰田研究院创建的大型数据集,包含1M个现实世界的视觉问答对。该数据集基于nuPlan,利用多视角图像和详细的注释生成自由形式的问答对,旨在支持驾驶场景的上下文感知分析。数据集被细分为九个子任务,涵盖三个核心技能:道路环境感知、空间关系识别和自我中心推理,以帮助多模态大型语言模型更好地理解和适应多视角驾驶场景。
NuPlanQA-1M is a large-scale dataset created by Purdue University and Toyota Research Institute, containing 1 million real-world visual question-answer pairs. Built upon nuPlan, this dataset generates free-form question-answer pairs using multi-view images and detailed annotations, aiming to support context-aware analysis of driving scenarios. The dataset is subdivided into nine subtasks covering three core skills: road environment perception, spatial relationship recognition, and egocentric reasoning, to help multimodal large language models better understand and adapt to multi-view driving scenarios.




