PSI

该数据集是从自然驾驶研究中抽取的110次车与行人遭遇的事件，对行人的行为、意图及推理过程进行了详尽的标注。该数据集不仅包含了视觉注释，还包含了认知注释，这使得在自动驾驶车辆交互背景下，能够对行人行为进行全面分析。该数据集规模涵盖了110次遭遇事件，超过62.1万个意图估计以及25,881个标注帧。其任务涵盖了行人意图/行为预测、车与行人交互分割以及视频到语言的映射。

This dataset contains 110 vehicle-pedestrian encounter events extracted from naturalistic driving studies, with exhaustive annotations covering pedestrian behaviors, intentions, and reasoning processes. Besides visual annotations, it also incorporates cognitive annotations, enabling comprehensive analysis of pedestrian behaviors in the context of autonomous vehicle interactions. In terms of scale, the dataset includes over 621,000 intention estimations and 25,881 annotated frames across these 110 encounter events. The supported downstream tasks cover pedestrian intention/behavior prediction, vehicle-pedestrian interaction segmentation, and video-to-language mapping.