Object-level Risk Assessment Dataset
收藏资源简介:
Object-level Risk Assessment Dataset是一个用于评估自动驾驶系统在感知受限区域中推理能力的大型多模态数据集。该数据集包含20万条问答对,涵盖了多种复杂的真实场景,特别是涉及动态和静态遮挡的场景。数据集的创建过程结合了GPT-4o和GPT-4o-mini模型,通过多步推理生成高质量的问答对,旨在评估模型在感知受限情况下的风险评估能力。该数据集主要应用于自动驾驶领域,帮助提升系统在复杂环境中的安全性和推理能力。
Object-level Risk Assessment Dataset is a large-scale multimodal dataset developed to evaluate the reasoning abilities of autonomous driving systems within perceptually constrained environments. It comprises 200,000 question-answer pairs spanning diverse complex real-world scenarios, particularly those involving dynamic and static occlusions. The dataset was created by leveraging GPT-4o and GPT-4o-mini models, generating high-quality question-answer pairs via multi-step reasoning, with the aim of evaluating models' risk assessment capabilities under perceptually constrained conditions. This dataset is primarily applied in the autonomous driving domain to help enhance the safety and reasoning capabilities of autonomous driving systems in complex environments.




