Commonsense Augmented VL Datasets

该数据集通过将现有的图像-文本对与从知识图谱中提取的常识知识相结合，特别是使用了ConceptNet进行增强。数据集的生成过程是自动化的，它将图像与文本描述中语言实体对应的常识知识进行配对。这一流程旨在提高视觉-语言模型中的常识推理能力。This dataset enhances existing image-text pairs by integrating commonsense knowledge extracted from knowledge graphs, with specific augmentation using ConceptNet. The dataset generation process is fully automated, which pairs images with commonsense knowledge corresponding to linguistic entities in their associated text descriptions. This pipeline is designed to improve commonsense reasoning capabilities in vision-language models.