ucf-crcv/ImplicitQA

ImplicitQA数据集是一个专门为测试视频问答中的隐式推理任务而设计的新型基准。它包含了1000个精心注释的问答对，这些问答对来自320个高质量的创意视频片段。与现有的主要关注显性视觉内容（如单个帧或短片段中的动作、对象、事件）的视频问答基准不同，ImplicitQA专注于模型在动机、因果关系以及跨非连续帧关系的推理能力。这个数据集模拟了人类对创意和电影式视频的理解，这些视频通常采用省略某些描述的故事讲述技巧。

The ImplicitQA dataset is a novel benchmark specifically designed to test models on implicit reasoning in Video Question Answering (VideoQA). It consists of 1,000 meticulously annotated QA pairs derived from over 320 high-quality creative video clips. Unlike existing VideoQA benchmarks that focus on questions answerable through explicit visual content, ImplicitQA addresses the need for models to infer motives, causality, and relationships across discontinuous frames, which mirrors the human-like understanding of creative and cinematic videos that often use storytelling techniques involving deliberate omissions.