VQA v1

该数据集是首个大规模、开放式的自由形式数据集，旨在评估视觉问答（VQA）方法的能力，如细粒度检测、识别和计数。它包含了614.2万个问题和204.7万张图片。此外，该数据集为问题提供了多个真实答案，并分为训练、验证、测试开发、标准测试、挑战测试和保留测试等多个部分。其规模属于大型，任务类型为视觉问答。

This dataset is the first large-scale, open-domain free-form dataset designed to evaluate the capabilities of Visual Question Answering (VQA) methods, such as fine-grained detection, recognition and counting. It contains 6.142 million questions and 2.047 million images. Additionally, the dataset provides multiple ground-truth answers for each question, and is divided into multiple subsets including training, validation, test-dev, standard test, challenge test and holdout test. As a large-scale dataset, its targeted task is Visual Question Answering.

搜集汇总

背景与挑战

背景概述

VQA v1是Visual Question Answering Challenge的第一版数据集，用于评估AI系统在回答关于图像的开放性问题上的能力。VQA v2.0是其改进版本，减少了语言偏见并增加了数据量。