QLEVR

QLEVR是一个用于量化语言和基础视觉推理的诊断数据集，由东北大学创建。该数据集包含100,000张合成图像和999,446个独特问题，旨在超越现有的存在性和数值量化，专注于更复杂的量化器及其组合。数据集创建过程中，通过自动构建场景图并生成合成图像，确保了图像中对象的位置、属性和关系的准确性。QLEVR的应用领域广泛，特别是在视觉问答系统中，用于测试和提高系统对复杂量化语言的理解和推理能力。

QLEVR is a diagnostic dataset for quantificational language and grounded visual reasoning, developed by Northeastern University. It comprises 100,000 synthetic images and 999,446 unique questions. Unlike existing datasets that focus solely on existential and numerical quantification, this dataset is designed to center on more complex quantifiers and their combinatorial applications. During its construction, scene graphs are automatically built and synthetic images are generated, ensuring the accuracy of object positions, attributes and relational details within the images. QLEVR has a wide range of application scenarios, especially in visual question answering (VQA) systems, where it is used to test and improve the systems' capabilities of understanding and reasoning about complex quantificational language.