IconQA (Icon Question Answering)

当前的视觉问答（VQA）任务主要考虑在日常生活环境中回答人工注释的自然图像问题。图标问答 (IconQA) 是一个基准测试，旨在强调抽象图表理解和全面认知推理在现实世界图表文字问题中的重要性。对于这个基准，构建了一个大规模的 IconQA 数据集，该数据集由三个子任务组成：多图像选择、多文本选择和填充空白。与现有的 VQA 基准相比，IconQA 不仅需要对象识别和文本理解等感知技能，还需要几何推理、常识推理和算术推理等多种认知推理技能。描述来自：IconQA

Current visual question answering (VQA) tasks primarily focus on answering manually annotated natural image questions within daily-life scenarios. Icon Question Answering (IconQA) is a benchmark designed to highlight the importance of abstract diagram understanding and comprehensive cognitive reasoning in real-world diagram-based textual questions. For this benchmark, a large-scale IconQA dataset has been constructed, which consists of three subtasks: multiple image selection, multiple text selection, and blank filling. Compared with existing VQA benchmarks, IconQA requires not only perceptual skills such as object recognition and text understanding, but also a variety of cognitive reasoning skills including geometric reasoning, commonsense reasoning, and arithmetic reasoning. Description source: IconQA

搜集汇总

背景与挑战

背景概述

IconQA是一个专注于抽象图表理解和认知推理的视觉问答数据集，包含三个子任务，强调多种认知推理技能。该数据集由多个知名大学和研究机构于2021年联合发布，提供了相关的论文和首页链接。