pengyueli/RAF_Argoverse

这是一个经过处理的RAF就绪版Argoverse 2数据集，专为支持检索增强的公平聚类框架（RAF）而设计。该数据集基于Argoverse 2，一个用于感知和预测研究的大规模自动驾驶数据集。处理后的版本包含查询数据集、外部源文件、源元数据、轨迹嵌入和敏感组标签，旨在解决自动驾驶领域中不同参与者组轨迹模式在嵌入空间分布不均的问题，通过数据层面的检索和选择来减少少数群体的代表性不足，以促进下游公平聚类任务。This is a processed RAF-ready version of the Argoverse 2 dataset, specifically designed to support the Retrieval-Augmented Dataset Assembly framework (RAF) for fair clustering. Based on Argoverse 2, a large-scale autonomous driving dataset for perception and forecasting research, the processed version includes the query dataset, external source files, source metadata, trajectory embeddings, and sensitive-group labels. It addresses the uneven distribution of trajectory patterns among different actor groups in the embedding space within the autonomous driving domain, aiming to reduce minority under-representation through data-level retrieval and selection for enhanced downstream fair clustering.