ZoeyLIU1999/EgoTraj-Bench

EgoTraj-Bench是第一个在自我中心噪声观测下进行行人轨迹预测的真实世界基准数据集。它基于TBD数据集构建，将噪声的第一人称视角（FPV）轨迹与干净的鸟瞰视角（BEV）地面真实轨迹配对，从而能够在部署现实的条件下对轨迹预测模型进行稳健评估。数据集分为三个级别：L0为原始数据，L1为中间处理数据，L2为可直接使用的处理后的数据。L2级别的数据包括EgoTraj-TBD和T2FPV-ETH两个部分，分别用于真实世界和模拟的自我中心噪声下的轨迹预测。数据集还提供了详细的统计信息和引用格式。

EgoTraj-Bench is the first real-world benchmark for pedestrian trajectory prediction under ego-centric noisy observations. Built upon the TBD dataset, it pairs noisy first-person-view (FPV) derived trajectories with clean birds-eye-view (BEV) ground truth, enabling robust evaluation of trajectory prediction models under deployment-realistic conditions. The dataset is structured at three levels: L0 for raw data, L1 for intermediate processing, and L2 for ready-to-use processed data. The L2 level includes two parts: EgoTraj-TBD for real-world ego-centric noise and T2FPV-ETH for simulated ego-centric noise. Detailed statistics and citation format are also provided.