EQA

该数据集名为EQA，包含了大约9000个问题，这些问题分布在774个为身体化问答任务而设计的环境中。在这些环境中，智能体必须在一个模拟环境中导航以回答视觉问题。数据集涵盖了与位置、颜色、房间内颜色和介词相关的四种类型的问题。值得注意的是，训练集、验证集和测试集之间没有任何重叠的环境。规模上，该数据集跨越了774个环境，共包含9000个问题，其任务是身体化问答。

This dataset, designated as EQA, contains approximately 9,000 questions distributed across 774 environments constructed for embodied question answering tasks. Within these environments, AI Agents must navigate a simulated scenario to respond to visual questions. The dataset includes four categories of questions related to location, color, in-room color, and prepositions. Importantly, there are no overlapping environments among the training, validation, and test splits of this dataset. In total, this dataset spans 774 environments with 9,000 questions, centered on the embodied question answering task.