ECHO

ECHO是一个基于视觉语言的社会场景事件因果推理诊断数据集，由新加坡国立大学创建。该数据集包含2000个推理实例，基于真实世界的人类中心推理信息，特别是电视犯罪剧集。ECHO要求具备心智理论（ToM）能力，以理解和推理多模态信息基础上的社会互动。数据集旨在评估当前AI系统在零样本和少量样本视觉语言推理中的能力，特别是在人类中心信息处理和因果推理方面的挑战。

ECHO is a vision-language diagnostic dataset for causal reasoning in social scenarios, created by the National University of Singapore. This dataset comprises 2,000 reasoning instances based on real-world human-centric reasoning information, particularly from television crime dramas. ECHO requires theory of mind (ToM) capabilities to understand and reason about social interactions grounded in multimodal information. This dataset is designed to evaluate the capabilities of current AI systems in zero-shot and few-shot vision-language reasoning, especially regarding the challenges of human-centric information processing and causal reasoning.