open-social-world/EgoNormia

EgoNormia是一个具有挑战性的问答基准测试，用于测试VLM模型在上下文中推理社会规范的能力。该数据集包含来自Ego4D的1853个基于物理的egocentric交互视频片段和对应的每个视频片段的五个选项多项选择题任务。EgoNormia涵盖了100个不同的场景，跨越了广泛的活动、文化和互动。与其它基于视觉的时空、预测或因果推理基准测试不同，EgoNormia评估模型在社交规范下应该做什么的推理能力。EgoNormia突出了这些与规范相关的目标冲突的情况，这是评估规范决策的最丰富的领域。

EgoNormia is a challenging QA benchmark that tests VLMs ability to reason over norms in context. The dataset consists of 1,853 physically grounded egocentric interaction clips from Ego4D and corresponding five-way multiple-choice questions tasks for each. EgoNormia spans 100 distinct settings across a wide range of activities, cultures, and interactions. Unlike other visually-grounded spatiotemporal, predictive, or causal reasoning benchmarks, EgoNormia evaluates models’ ability to reason about what should be done under social norms. EgoNormia highlights cases where these norm-related objectives conflict—the richest arena for evaluating normative decision-making.