InfoSeek

该数据集专为那些不能仅凭常识知识回答的信息寻求类问题而设计，是一个视觉问答数据集。它突显了模型在回答视觉信息寻求问题时所面临的挑战，并展示了通过微调可以从预训练模型中引出知识。该数据集的任务是视觉问答。

This is a visual question answering (VQA) dataset designed specifically for information-seeking questions that cannot be answered solely based on common sense knowledge. It highlights the challenges models encounter when addressing visual information-seeking queries, and demonstrates that knowledge can be elicited from pre-trained models via fine-tuning. The task of this dataset is visual question answering.

搜集汇总

背景与挑战

背景概述

InfoSeek是一个专注于视觉信息寻求问题的VQA数据集，包含人工标注的高质量问答对和自动构建的大规模数据，用于评估预训练视觉问答系统的性能。