GuessWhat?!

该数据集是一个基于两人猜谜游戏的大规模数据集，包含了150,000场由人类玩家参与的游戏，产生了800,000组关于66,000张图片的视觉问题与答案对。此外，该数据集有助于提升对话系统的交互能力，其任务重点在于基于图像的对话以及空间推理能力。

This is a large-scale dataset based on a two-player guessing game. It includes 150,000 games played by human players, generating 800,000 visual question-answer pairs corresponding to 66,000 images. Furthermore, this dataset contributes to enhancing the interactive performance of conversational systems, with its core tasks focusing on image-based dialogue and spatial reasoning capabilities.