OpenDCAI/dataflow-mm-context_vqa

DataFlow-MM-ContextVQA是一个大规模合成的多模态数据集，包含超过200,000个视觉问答（VQA）实例。每个实例都包含一个图像、一个自然语言问题和一个相关的上下文文档，该文档包含回答问题所需的信息。数据集的设计目的是强调上下文感知的多模态推理，要求模型同时利用视觉内容和外部文本上下文来准确回答问题。所有样本都是在DataFlow框架内自动生成的，确保了一致性和可扩展性。

DataFlow-MM-ContextVQA is a large-scale synthetic multimodal dataset consisting of over 200,000 visual question–answer (VQA) instances. Each example pairs an image with a natural language question and an associated context document that contains the information required to derive the correct answer. The dataset is designed to emphasize context-aware multimodal reasoning, where models must jointly leverage visual content and external textual context to answer questions accurately. All samples are automatically generated within the DataFlow framework, ensuring consistency and scalability.