GQA

该数据集基于Visual Genome的场景图开发而成，包含了具有单一正确答案的复杂问题，总计有2200万个问题和113,000张图片。此外，该数据集还为问题-答案对提供了详细的视觉注释和推理步骤。它是一个大规模的数据集，适用于视觉问答任务。

This dataset is developed based on the scene graphs of Visual Genome, containing complex questions with a single correct answer, totaling 22 million questions and 113,000 images. Furthermore, it provides detailed visual annotations and reasoning steps for each question-answer pair. As a large-scale dataset, it is suitable for visual question answering (VQA) tasks.