Visual Dialog

Visual dialog is a multi-round extension for VQA. The interactions between the image and multi-round question-answer pairs (history) are progressively changing, and the relationships among the objects in the image are influenced by the current question.

视觉对话（Visual Dialog）是视觉问答（Visual Question Answering, VQA）的多轮扩展范式。图像与多轮问答对（即对话历史）之间的交互会逐步动态演变，而图像中各目标间的关联关系亦会受到当前问题的影响。