PaSBench

该数据集名为PaSBench，旨在评估多模态情境下的前瞻性安全意识，涵盖了416个场景，其中包括128个图像序列和288个文本日志，这些场景分布在五个安全关键领域。该数据集将风险分为五个领域：家庭、户外、运动、食品以及自然灾害和紧急情况。它通过人工参与的方式构建，专注于特定用户的风险。数据集的规模包括416个场景（128个图像序列，288个文本日志），其任务是基于文本或图像观察进行前瞻性风险检测。

The dataset named PaSBench is designed to evaluate proactive safety awareness in multimodal contexts. It encompasses 416 scenarios, comprising 128 image sequences and 288 text logs, which are distributed across five safety-critical domains. This dataset classifies risks into five categories: household, outdoor, sports, food, as well as natural disasters and emergencies. It is constructed via human involvement, focusing on risks pertaining to specific user groups. The core task of PaSBench is to perform proactive risk detection based on textual or visual observations.